Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure, and industrial technology. This week we look at China’s EV subsidies, the future of aeroderivative gas turbines, a US-Saudi nuclear deal, defective roadway guardrails, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housing and Cities

The CEO of DR Horton, largest homebuilder in the US, thinks that the 21st Century ROAD to Housing Act won’t have much impact on housing supply or demand in the short term. [X]

The Financial Times on why socialists should embrace luxury apartments. “Above all, high-end developments unlock long housing chains. As higher-income households move into newly built units, they free up older properties, raising supply and slashing prices for middle- and lower-end housing through a process known as filtering. Numerous international studies underscore this positive ripple effect.” A good sentiment, though stating it to the readers of the Financial Times is probably preaching to the choir. [FT]

Last week we noted that Saronic, which had been slated to be a major tenant for California Forever’s shipbuilding facility, had instead opted to locate their operations in Texas. Someone claimed on Twitter that California was, in fact, never really in consideration, and that overtures to California Forever were entirely to secure better terms in Texas. [X]

The fraction of housing markets in the US where prices are rising vs falling over time, via ResiClub. [X]

US cities which used to have at least 100,000 people but no longer do. [Reddit]

Manufacturing

TSMC is accelerating its investment in Arizona fabs due to enormous AI demand. [CNBC] And TSMC plans to raise prices by up to 10% starting in 2027. [Yahoo Finance]

As China’s EV industry gets more mature, China is beginning to roll back subsidies for the sector. “Starting Sept. 1, lithium-ion batteries will be subject to a 2% consumption tax, which will double to 4% a year later, according to a Friday announcement by the Ministry of Finance.” [X]

Energy

The National Historic Preservation Act (NHPA) requires considering whether a new project will impact the view of some existing historic property. Thus the more land area a project extends over, or the taller it is, the more potential view impacts, and the greater the NHPA review burden. This is bad news for clean energy projects like wind and solar, which have a much “viewprint” than oil and gas projects. “We examined nearly 100 recent energy projects and found a striking discrepancy: compared to fossil energy projects, clean energy and transmission projects have review areas about 10 times larger and are more than 8 times as likely to receive an “adverse effect” finding — triggering additional process, delay, and mitigation obligations.” [IFP]

Tim Latimer, CEO of geothermal startup Fervo Energy, on what the geothermal industry can learn from SpaceX. Basically, instead of trying to develop an ambitious technology in one shot, start with a simpler, less ambitious version of it, and advance via learning by doing. “...much like the Starship example, starting with the end state and trying to solve all the various technology challenges all in one go would be incredibly challenging. So Fervo has opted to take a learning by doing path that parallels SpaceX. Project Red was our first pilot. Rather than try to drill a super hot or super deep concept right out of the gate, we completed a project in a category called “near-field Enhanced Geothermal Systems” (NF-EGS). This meant going to a location that was adjacent to a producing geothermal site, so the geothermal gradient was higher, even though the permeability in the part of the field we had selected did not support prior development with conventional technology.” [LinkedIn]

Read more

Last time in our series on the nature of the construction productivity problem, we looked at economies of scale, finding that capturing economies of scale in homebuilding is difficult. Economies of scale primarily operate on the difference between the costs of the material inputs to some process and the cost of the final product. With housing construction, this ratio is already pretty low, not much higher than what we see in high-volume manufacturing industries that already maximize these kinds of cost advantages, such as the auto industry.

There is, however, another possible route for reducing construction costs: reducing the costs of the inputs directly, either by using cheaper inputs (less expensive materials or labor) or by using fewer inputs. To use a baking analogy, capturing economies of scale is like improving the efficiency of baking cakes by making them in a high-volume, industrial bakery instead of your home kitchen. Reducing the input costs is more like changing the recipe of that cake to require cheaper/fewer ingredients.

For housing construction, we can broadly categorize the inputs to construction as either materials or labor. (There are also equipment/machinery costs, but these are a very small fraction of housing construction costs.) Reducing the cost of labor inputs is difficult, for the simple fact that conventional construction is done on-site, limiting the ability to use cheaper pools of labor. Reducing the cost of material inputs appears somewhat more viable in theory (at least for some materials), but a variety of barriers make actually doing so difficult in practice.

Finding cheaper construction labor is hard

Historically, manufacturers have reduced their labor costs by moving their operations to places where labor is cheaper. As I note in “The Origins of Efficiency”:

Garment manufacturing, for example, has proven resistant to automation and remains a labor-intensive industry. As a result, garment manufacturing operations tend to continuously relocate to sources of low-cost labor (which are often, not unrelatedly, places with poor working conditions). By the late 19th century, New York City had become one of the largest garment manufacturing centers in the world, such that in 1890 New York made 44 percent of the ready to wear clothes produced in the US. But by the 1920s, the industry had begun to move where labor was cheaper — first from Manhattan to Brooklyn and New Jersey, then to New England, then, as the interstate highway system developed, to the South, and, finally, overseas. Similarly, Nike began importing footwear from Japan in the 1960s, then began manufacturing its own shoes in Japan in the 1970s. As labor costs in Japan rose, Nike shifted production operations to Taiwan and Korea, then China, then Vietnam and Indonesia. There have been similar shifts in other labor-intensive industries, such as shipbuilding (which moved from the UK to Japan, then to South Korea, then to China), as well as toy manufacturing (which moved from the US to Japan, then to Hong Kong, then to China).

This strategy, however, is difficult to pursue in construction, for the obvious reason that conventional construction is done on-site and can’t be relocated to where labor happens to be inexpensive.

A relocation strategy does become available if your construction is prefabricated, but this introduces new complications, notably the high costs of transporting prefabricated building components. As we’ve previously noted, most prefabricated construction is produced in distributed, relatively small-volume factories located to minimize transportation distance. (A common limit, noted by several manufactured home producers, is 500 miles, about the maximum that a truck can drive in a single day.)

There are labor cost differentials in the US, and modular builders do indeed locate their operations to take advantage of them. If we use RSMeans City Cost Index values for “installation costs” as a proxy for construction labor costs, there’s a roughly 2x cost difference between the 10th percentile city and the 90th percentile city.1 Modular builder Autovol fabricates its modules in Nampa, Idaho, and ships them hundreds of miles (far beyond the typical 500-mile limit) to Los Angeles and the Bay Area, where labor costs are far higher. Stack Modular builds its modules in China and then transports them to West Coast building projects using bulk carriers. Volumetric Building Companies has similarly noted that modular construction works best when a large labor cost differential exists between the factory location and the site location. But the added capital and transportation costs of modular construction dull this benefit. Autovol does boast that its construction costs are lower than conventional construction in the high-cost California metros that it builds in. But Stack Modular, despite using low-cost Chinese labor, only makes the more modest, more common claim of the benefits of prefabrication, that it provides “cost certainty” and “reduced schedule” rather than lowering hard costs. Volumetric Building Companies similarly notes that the cost savings of prefabricated construction, when they exist, are typically modest (5–10%), and often don’t appear at all.

This constraint of being unable to relocate to a source of low-cost labor is shared by the agriculture industry, which likely explains why construction and agriculture are the two industries with the largest shares of (presumably lower-cost) undocumented labor.

Another strategy for using cheaper labor is to rework your production methods to allow the use of less skilled, and thus less expensive, labor. This was a major advantage of Henry Ford’s mass-production methods; the special-purpose machine tools Ford developed typically did not require a skilled machinist to operate. (Ford called the devices he used to streamline production “farmer’s tools,” because they would allow a farmer to produce parts as well as a trained mechanic.) But doing this requires some combination of sufficient economies of scale to amortize the equipment costs (which we’ve previously established as difficult) and changing the construction technology employed. We’ll look at the potential of this latter option in a future essay.

Lowering material costs is hard

The other strategy for reducing input costs is to lower material costs, either by using less expensive materials or using fewer materials.

Finding a way to use fewer construction materials, often for the perceived environmental benefits rather than cost savings, is a common strategy for would-be developers of new building systems. One of the supposed benefits of concrete 3D printing, for instance, is that it allows for much more materially efficient construction by placing concrete only where it’s needed.

The problem with this strategy, however, is that most existing construction is likely fairly close to the “material efficiency” frontier. As I noted in a previous essay on reducing the material necessary for building, the entire job of structural engineers is already to minimize the size of the structure needed to support a given load:

Structural elements are designed to be as materially efficient as possible, and have been for many decades. Wood trusses (made up of conventional lumber stitched together with steel plates) and wood I-joists (made up of engineered lumber top and bottom chords with a center “web” of oriented strand board, or OSB) are standard for residential construction in the US. Steel W-sections and steel joists (basically trusses) are standard for commercial construction. For even lighter steel construction, there is light gauge steel framing, which consists of very thin sheets of steel bent into structurally efficient shapes. There are also things like insulated metal panels, which are in some ways the platonic ideal of an efficient structural element: a thin sheet of steel on the top and bottom, separated by a layer of insulation. (The wood version of insulated metal panels, SIPS, never quite caught on in the US, though not for lack of trying.)

Even heavy, cheap materials like concrete are used in materially efficient ways. For concrete floors, a common construction method is to pour concrete over a ribbed metal deck. The ribs give the steel the strength to support the concrete while it’s wet (eliminating the need for additional shoring), and give the whole assembly greater depth, increasing bending resistance and placing the steel at the outer edge where it’s most useful. There are also things like hollowcore slabs (concrete slabs manufactured with voids in the middle), and concrete blocks (which are hollow in the middle, though many of the voids will have reinforcement placed in them and be grouted solid).

When structural material use isn’t minimized, it’s often because further reductions trade off against something else, such as higher labor costs, deeper and more expensive floor depths, or increased construction complexity. There are many structural systems, such as castellated beams (beams with the centers cut out), advanced framing (house framing designed to minimize the use of dimensional lumber), or concrete void slabs (concrete floors cast around hollow plastic spheres), that are rarely used because they require more labor, have more complexity (and thus carry more risk), or have some other undesirable tradeoff.

Another barrier that can prevent the reduction in building materials is code requirements. Requirements for electrical wiring, for instance, were set assuming the use of relatively power-hungry incandescent lighting. With the rise of LED lighting, there’s an opportunity to reduce wiring requirements for lighting by using low-voltage DC wiring; some startups have been founded to try and tackle this, but the use of lower-voltage wiring has been limited by building code restrictions.

If using fewer building materials isn’t an obviously winning path, what about using less expensive building materials? Here there do appear to be somewhat more opportunities, at least theoretically, though capturing them is far from straightforward.

Many building materials, to be sure, are poor candidates for either cost reduction or replacement with a cheaper substitute. The largest single line item when constructing a new house, for instance, is the structural framing, which, with labor and materials, makes up around 20% of the cost of a new house. But dimensional lumber is already exceptionally inexpensive: in dollars-per-cubic-foot terms, it’s among the least expensive materials that civilization produces.

If we look at the gross margins of dimensional lumber producers, they’re not particularly high: Boise Cascade had a gross margin of 16.5% in 2025, when Weyerhaeuser had a gross margin of 14.8%. It’s thus not straightforward to either drive down the cost of wood framing or substitute it with a cheaper material.2

Similarly, there’s historically been a great deal of interest in finding an alternative to drywall, but virtually every alternative is significantly more expensive:

Despite its drawbacks, installing drywall is incredibly inexpensive. My 2022 Construction Estimator gives a cost of installing + finishing drywall at about $1.50 per square foot. RSMeans gives similar. Most of these alternative materials are much more expensive, even before taking into account the labor or extra backing materials they might require. MDF panels seem to be ~$1.50 for the material alone. PVC panels seem to be in the realm of $3.50 per square foot for just material. Fibo is closer to $10 per square foot, Corian around $50 per square foot, and Dekton up to $100 (though this is for the countertops, walls might be cheaper).

Other materials, however, seem like they have more opportunity for cost reduction, though these reductions aren’t necessarily enormously large or easy to capture. Concrete, for instance, is not particularly expensive (typically costing ~$6 per cubic foot), but if you look at the gross margins of concrete material suppliers, they’re often surprisingly high. Vulcan Materials, which makes aggregate, had a gross margin of 27% in 2025. Martin Marietta had 31% that year, and Cemex had 32%. These are high enough that you could imagine a modest decline in the cost of concrete in a more competitive concrete industry.

But it’s hard to get that increased competition, in large part because it has become extremely difficult to permit a new aggregate quarry. In some parts of Washington, it can take 7 to 10 years and upwards of $1 million to permit a new quarry, and the prospect is so risky that producers are often unwilling to even try. In California, Granite Construction spent 7 years trying to open a quarry, including writing an 8,500-page environmental impact report, only for the permit to be denied. For cement, the story is similar: St. Lawrence Cement spent 7 years trying to build a new cement plant in New York before giving up in 2005. There hasn’t been a new greenfield cement plant built in the US in 15 years. Without the ability to build new quarries or cement plants, we can’t expect prices to come down.

There also seems to be room for lower material prices for steel: prices for American steel are substantially higher than for European or Chinese steel. But this also isn’t straightforward to address. Tariffs and transportation costs whittle away the cost savings of importing foreign steel, and starting a new, more efficient American steel producer doesn’t seem like an obvious win, given the already-low gross margins of US steel producers even with the tariffs (12% for Nucor, 15% for Commercial Metals Co., and negative for Cleveland-Cliffs in 2025).

Similarly, many other building product suppliers have high enough margins that major cost reductions seem like they should theoretically be possible. James Hardie Industries, which manufactures fiber cement siding, had a gross margin of nearly 36% in 2025. Owens Corning, which produces insulation, roofing, and doors, had 28%. Armstrong World Industries, a manufacturer of ceiling and architectural products, had 40%. Simpson, a manufacturer of structural connectors, had a gross margin of 46%.

But as with the other materials we looked at, there are barriers to the increased competition that might drive down these costs. For many of these products, it would often be risky for engineers, architects, or building designers to specify an alternative. These products have typically gone through extensive semi-official testing, documented in things like ICC or IAPMO code reports, which give designers a third-party verification of what level of performance they can expect from the product. Products are often backed with warranties, and the companies typically put a great deal of effort into making it easy for designers to specify their products, through things like well-designed catalogs, free design software, and free engineering assistance. (Simpson’s connector catalog, for instance, is so well put together that it was actually used as a textbook in a course I had on wood design in college.)

Building professionals, as I’ve noted previously, are rationally risk-averse: the upside from specifying a slightly cheaper product is relatively minor compared to the huge downside risk of a building product that doesn’t work as expected. When a building product doesn’t work as advertised — polybutylene piping in the 1980s, EIFS cladding in the 1990s — engineers, architects, and builders are invariably drawn into the resulting lawsuits. This makes it hard to compete with existing manufacturers of many products: most engineers are reluctant to specify specify some no-name brand of connector over Simpson connectors, when a connector failure could cause a catastrophic failure in an earthquake or hurricane.

The construction startup Katerra (where I used to work) found this out the hard way. When it was first founded, it was a supply chain and logistics company. It would source low-cost materials from China and elsewhere and offer them more cheaply than competitors to builders in the US. But this plan didn’t work, because architects and builders weren’t willing to specify Katerra’s products, and Katerra pivoted to prefab construction, using factory methods and sourcing its own materials in the hopes of outcompeting other builders.

We also see the same sort of tradeoff problem at work with material substitutions as we did with material reductions. Often it’s possible to replace an expensive material with a cheaper one, but this incurs a tradeoff that people are often unwilling to make, particularly if the product is an interior finish. Fiberglass bath enclosures, for instance, are both less expensive and higher performance (i.e., less likely to leak) than ceramic tile, but tile is almost always preferred for aesthetic reasons. Similar aesthetic preferences exist for things like solid vs. hollow doors, granite vs. Formica countertops, fiber cement or brick vs. vinyl siding, drywall vs. vinyl-on-gypsum panels with visible seams, and so on. Cheaper finish materials often feel cheaper, either because they’re thinner and feel less solid or simply because we can tell the difference between them and more expensive materials.

Conclusion

Overall, prospects for lowering construction costs by way of finding a cheaper “recipe” — cheaper or fewer ingredients like materials or labor — are something of a mixed bag. For labor, the possibilities of cost reduction seem small, given construction’s on-site nature and the difficulties associated with off-site, prefabricated construction in low-labor-cost locations. For building materials, the possibilities are somewhat greater. While using more materially efficient designs likely has relatively little in the way of opportunity, many building materials seem like they could theoretically be made less expensively (either in the US or overseas). But a variety of barriers — the difficulty of permitting new production facilities, tariffs keeping out low-cost foreign supplies, the risk aversion of architects, engineers, and building designers — make achieving this less than straightforward.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at a setback for California Forever, a $500 million artillery shell plant not producing any shells, the difficulty of recycling EV batteries, manufacturing wire harnesses, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housekeeping items this week:

• Origins of Efficiency is apparently one of the most purchased books by startup founders, according to Brex (a credit card aimed at startups).

Housing and Cities

We’ve previously seen homeowners in the Bay Area asking for OpenAI or Anthropic stock when they sell their house, but apparently you also see landlords asking for equity share as part of the leasing agreement. (I’ve seen some claims this is fake, but Byrne Hobart noted that this sort of deal took place with Paypal.) [X]

A setback for California Forever, which is trying to build a new California city northeast of the Bay Area. Part of the plan involved a huge new shipyard, but one of the tenants for that shipyard, automated vessel startup Saronic, is now setting up shop in Texas instead. [SF Chronicle]

California Forever’s bid to recruit a major industrial tenant for a planned waterfront shipbuilding facility in Solano County has slipped away, a significant blow to the ambitious but heavily scrutinized billionaire-backed plan to build an entire new city on thousands of acres of farmland.

Defense startup Saronic Technologies, which designs and builds autonomous water crafts, has chosen Texas — where the company has been considering the Port of Brownsville — over Solano County for its $3.2 billion Port Alpha automated shipyard, according to multiple sources familiar with the negotiations.

In my essay about build to rent housing, I noted that one developer I spoke to said that one benefit of rental housing was giving access to good school districts to people who couldn’t otherwise afford to live there. A recent paper in Real Estate Economics suggest this is actually a pretty significant benefit (though I haven’t read this paper closely). [Wiley]

Using a unique data set of linked student-housing-unit records from North Carolina and a fixed effects research design, we document an association between increases in the supply of single-family rental units (SFRs) and the likelihood that economically disadvantaged children attend higher performing schools. Drawing on exogenous variation in the allocation of SFRs across neighborhoods, we find that children in renter households were more likely to experience improvements in school quality when relocating to areas with an ample supply of SFRs zoned for high-performing schools. Using a shift-share identification strategy with school-level data, we also provide evidence supporting the generalizability of our results to other markets throughout the United States.

The Wall Street Journal on how Baby Boomers, unlike their parents, aren’t necessarily downsizing to smaller homes when they retire. [WSJ]

Manufacturing

A cool video of automated carbon fiber manufacturing. [X] And a cool video of Tesla ripping out the Model X/Model S production line. [X]

The New York Times on the enormous setbacks in US EV manufacturing. Manufacturers pivoted away from them so quickly that in one case Chevy cancelled production of an EV between when they gave a review model to a publication and when that review came out. “In purely financial terms, the combined cost of this industry about-face remains nothing short of staggering: This year, Stellantis alone was forced to write down $26 billion in E.V.-related losses. (Ford reported a slightly less ghastly $19 billion loss.) But somehow, it’s the long-term repercussions that look worse. “The way I’d put it,” the auto journalist Martin Padgett told me recently, “is that we pulled a U-turn while the rest of the world was pushing forward.” [NYT]

Related, things look pretty bleak for US EV manufacturer Lucid Motors. “Lucid Group on Tuesday denied as “completely false” a blog ‌post saying it was considering a potential take-private transaction or a Chapter 11 bankruptcy filing, after the electric-vehicle maker’s shares tumbled more than 50% in what would be their steepest one-day decline [BP: though they ended the day down 16%, not 50%]. Lucid said it had sufficient ​liquidity to fund operations well into the next year, and had not formed a special board committee to ​explore the reported scenarios. It also said restructuring adviser AlixPartners was assisting the company on improving ⁠execution and operations, and was not recommending bankruptcy.” [Reuters]

A $500 million plant for producing 155mm artillery shells has yet to produce any, despite coming online more than two years ago. The culprit seems to be the fact that General Dynamics tried to repurpose the original shell manufacturing equipment, designed for a different type of artillery shell, rather than spin up an entirely new production line. [The War Zone]

Read more

None of these effects of NEPA, however, were envisioned when the law was written. NEPA was seen primarily as an (aspirational) statement of US environmental policy, which was to “encourage productive and enjoyable harmony between man and his environment, to promote efforts which will prevent or eliminate damage to the environment and biosphere and stimulate the health and welfare of man; [and] to enrich the understanding of the ecological systems and natural resources important to the Nation.” The provision that requires environmental impact statements was added last minute as a way to try to give some teeth to these high-minded but somewhat abstract ideals, and received virtually no attention at the time. As Alec Stapp and I noted:

[The provision] was not covered in any major media publication. In Congress, it received “neither debate, nor opposition, nor affirmative endorsement.” Caldwell would later state that “most [members] had never really understood the bill and only agreed to it because it was from Jackson; it was about the environment which was a very ‘hot’ issue at the time; and it was almost Christmas and they wanted to get home.”

Not until several months after NEPA was passed did environmental groups realize what a potent weapon they’d been handed.

It’s not hard to find other examples of laws whose effect was far different than what the authors anticipated. The 401(k) retirement account, now used by tens of millions of Americans as the primary vehicle for retirement savings, was originally considered an insignificant provision of the 1978 Revenue Act. Per a Bloomberg piece about the Act:

…The initial provision was estimated to have a “negligible effect upon budget receipts.” Now, defined contribution plans are the fifth-biggest tax break for individuals, with an estimated revenue loss to the government of $61.4 billion in fiscal 2014.

“There was absolutely no discussion in ’78 that if you do this, the world is going to change,” said Daniel Halperin, then a senior Treasury official and now a Harvard Law School professor.

The tale of Richard Stanger [a primary author of the Act], who said he hadn’t been interviewed previously about his role, is also a story about accidental actors at historic moments. As Stanger himself says, if anyone had known how important 401(k) would become, the Joint Committee on Taxation never would have let him, a 28-year-old junior lawyer, write it.

In the other direction, laws aimed at stimulating the construction of housing in California have proven much less effective than predicted:

One California law was supposed to flip defunct strip malls across California into apartment-lined corridors. Another was designed to turn under-used church parking lots into fonts of new affordable housing. A third would, according to supporters and opponents alike, “end single-family zoning as we know it.”

Fast-forward to 2025 and this spate of recent California laws, and others like it intended to supercharge the construction of desperately needed housing, have had “limited to no impact on the state’s housing supply.”

That damning conclusion comes from a surprising source: A new report by YIMBY Law, a pro-development nonprofit that would very much like to see these laws work.

I wanted to better understand how common this was: how often do laws do more or less what they’re designed to do? How often do their effects diverge widely, either by having unanticipated effects or by failing to do what the authors predicted? So I used AI to analyze the effects of several hundred federal laws passed over the last 50 years.

Overall, I found that federal laws mostly do what they’re expected to do. But a substantial fraction of them — around 11% — diverge significantly, having either much smaller or much larger effects than originally predicted.

Method

To do this analysis, I first chose five random federal laws passed each year from 1976 to 2023, filtering out any laws that were less than 10 pages in length, which were mostly insignificant things like post office renamings. This yielded 240 laws total, but for one law the AI was unable to find any information, so the actual analysis was done on 239 laws. For each law, I had an AI model — Claude Opus 4.8 Max Thinking — estimate the expected effect of the law, its actual effect, and assign a score to the divergence. Divergence scores ranged from -10 to 10: positive numbers indicate the law had a larger effect than anticipated (such as the 401(k) provision in the 1978 Revenue Act or NEPA), while negative numbers indicate the law had a smaller effect than anticipated.

Scoring Rubric:

• 0 — actual impact matched expectations essentially exactly.

• ±1–2 — minor: broadly as expected; small deviations on secondary dimensions that didn’t change the essential outcome.

• ±3–4 — moderate: clearly noticeable gaps on one or more dimensions, but the core purpose was still substantially realized (or missed only in a limited way).

• ±5–6 — substantial: the primary expected outcome was materially exceeded (+) or unmet (−), or a significant unanticipated effect emerged.

• ±7–8 — major: the central goal greatly overshot (+) or largely failed / went unimplemented (−), or effects were largely of a different kind than intended (+).

• ±9–10 — extreme: actual impact bore little resemblance to expectations — dominated by unintended, larger-than-anticipated, or opposite effects (+), or near-total failure / non-implementation (−).

There were a few complications in deciding how the effect of laws should be evaluated. One is deciding when a law should get credit for having an effect. Often the largest effect of a law only happens when subsequent laws extend, modify, or build on the original law. For instance, an unanticipated effect of the 1978 Psychotropic Substances Act was the introduction of civil asset forfeiture for drug proceeds. This mechanism became a cornerstone of US drug enforcement, but much of this later expansion was due to the 1984 Comprehensive Crime Control Act. For these cases, I instructed the AI to give some credit to the original law if it was built on by other laws, but to temper it based on how much subsequent laws actually did the heavy lifting. (For the Psychotropic Substances Act, the AI assigned it a score of +4, a moderate unanticipated effect, since the later bill did most of the work.)

Another complication was trying to determine what the expected effect of a law was at the time it was passed. For this, I instructed the AI to only use contemporaneous sources, such as the bill text itself, the presidential signing statement, a CBO score, press coverage at the time, etc. But this is inherently a fraught exercise: it’s often not obvious, for instance, to what extent the goal of some law is aspirational that the authors don’t expect to necessarily happen. For instance, the Indoor Radon Abatement Act of 1988 states in the bill text a national long-term goal that “indoor air in a building be as free of radon as the ambient outdoor air.” This is almost certainly an aspirational goal that the authors did not expect the bill to actually achieve, but making these judgments requires a process of guessing and using context clues that is likely to be error-prone. (The AI scored this law as a -2, a minor shortfall compared to what was expected; the various anti-radon programs the law created stuck around, but indoor radon exposure did not improve, in part because the programs were almost all voluntary.)

This exercise is similar to a previous post where I used AI to try and estimate how early various inventions could have been invented, but this turned out to be far more difficult and annoying, mostly because of the research required. For the previous exercise on inventions, I simply relied on the AI’s knowledge of various inventions to make the judgments. But for this present effort about laws I needed the AI to thoroughly research each law: I couldn’t expect the AI to know, word-for-word, every esoteric law passed in the last 50 years, or the complete history of the downstream effects of that legislation. I ended up needing to do a fair amount of trial and error to get Claude to do a sufficiently thorough job evaluating the expected and actual effects. I kept having to modify the prompt to force increasing levels of thoroughness in the search, and even in the final version I settled on I was occasionally finding effects that the AI evaluation had missed. Because of this, I expect there to be errors in various evaluations, and I would regard these results as preliminary.

You can see the full prompt I used, and the resulting evaluations, here.

Results

The graph below shows the divergence scores of the 239 evaluated laws.

The results form something like a bell curve with a slight positive skew. Ninety-five of the 239 laws (40%) had a score from -1 to 1: either they behaved exactly as expected or had very slightly larger or smaller effects than predicted. Forty-nine had a score from -2 to -4 (20%), and 68 (28%) had a score from 2 to 4, a moderate divergence. Only around 11% of laws had a “substantial divergence,” a score of plus or minus 5 or more. Most federal laws, it seems, do more or less what they’re expected to do.

(The graph above shows scores bundled together, but if you look at frequencies of individual score values, you don’t get a smooth bell shape. Instead you get a dip, with many fewer scores at -1 and +1. This is likely an artifact of the scoring rubric, which probably pushed scores either into 0 or out to plus or minus 2, but it’s another reason why we should treat these results with a grain of salt.)

Some examples of laws that Claude scored as having a much higher effect than predicted:

“An act to amend title XIII of the Federal Aviation Act of 1958 to expand the types of risks which the Secretary of Transportation may insure or reinsure, and for other purposes”, score +5: This mundane-sounding law is described by its title as mainly about airline insurance, but one of its provisions deregulated airline cargo service, the first step in deregulation of the airline industry more broadly. This deregulation was expected to increase competition and efficiency in the air cargo market, but its effect went beyond that. With air cargo flights deregulated, companies like FedEx, which were previously confined to using very small aircraft, could now use large jets on any route they wanted, setting the stage for the entirely new “air express cargo” industry.

The Sarbanes-Oxley Act of 2002, score +5: This law, passed in the wake of business collapses like Enron and WorldCom, was aimed at restoring investor confidence by increasing financial auditing oversight, and creating stiffer penalties for compliance failures. The additional oversight was achieved, but at a cost: compliance costs were 30 to 50 times higher than expected. Another positive divergence came from the fact that one of its clauses, which penalizes “obstruction of an official proceeding,” was later used in an unexpected way: criminally charging hundreds of January 6 defendants (though this was later struck down by the Supreme Court).

The Trade Facilitation and Trade Enforcement Act (TFTEA) of 2015, score +7: This law was billed as a customs modernization and trade enforcement act: giving more resources to enforce trade agreements, streamlining various trade and customs regulations, etc. At the signing ceremony, President Obama described it as “making sure other countries are playing by the rules.” But one of the provisions of the law changed the “de minimis exemption” — the value below which imported goods were not subject to tariffs — from $200 to $800. This change is credited as a driver of the explosion of Chinese imports from companies like Shein/Temu over the next several years, until it was reversed by the second Trump administration.

And here are some examples of laws that had a much smaller effect than predicted:

The Alaska Natural Gas Transportation Act of 1976, score -7: This act was expected to create a huge 4,800-mile pipeline, the “largest privately financed energy project ever undertaken,” that would transport natural gas from Alaska to the lower 48 states. The pipeline, however, was never completed, due to a combination of rising costs and the later Natural Gas Policy Act and Fuel Use Act creating a gas supply glut, obviating the economic justification for the pipeline.

The Alabama-Coosa-Tallapoosa River Basin Compact of 1997, score -5: This law was intended to create a commission that would develop a plan to share the water of the Alabama-Coosa-Tallapoosa River Basin between Alabama and Georgia. But while the commission was formed, Alabama and Georgia never agreed on an allocation formula, resulting in continuous litigation between the two states over water distribution.

The Enhanced Partnership with Pakistan Act of 2009, score -5: This act was expected to foster a closer relationship and increase goodwill with Pakistan, by providing billions of dollars’ worth of funding for schools, roads, and other infrastructure projects. This didn’t occur: opinion of the US in Pakistan continued to fall following things like President Obama visiting India but skipping Pakistan in 2010 and the US’s raid on Bin Laden’s Pakistan compound in 2011. By 2012, 74% of Pakistanis viewed the US as an enemy.

Other than the fraction of laws with significant divergence, there are a few other notable patterns in the data. If we look at divergence over time, we don’t see much change: recent laws seem roughly as predictable as older laws.

What about differences between small/minor laws and large/major laws? If we graph a law’s divergence score against the number of pages in the law, we see a small positive correlation: large laws with many pages are somewhat more likely to have a larger-than-expected effect than laws with fewer pages.

One possible explanation here is a sort of bundling effect: major laws, like the recent 21st Century ROAD to Housing Act, are often amalgamations of many smaller laws. Because the distribution of divergences is somewhat positively skewed, when you bundle many laws together, the chance that at least one of them has a large positive effect might rise.

Another notable pattern is that different types of laws have somewhat different probabilities of a major divergence. Appropriations bills, for instance, which are often (but not always!) routine allocations of money, have a lower variance than bills that create substantially new programs. Both types of bills have the same average divergence score, but the probability of a large or small divergence is much greater with the latter than the former.

Why do laws diverge from what’s expected?

At a high level, the reasons that laws have greater or smaller effects than expected can be divided into two categories: operators of the legal machinery behaving differently than expected, and the broader world, including those who the law was designed to affect, behaving differently than expected.

On the “legal operator side,” this includes everyone who has a role in authoring, enforcing, or interpreting laws. Courts, for instance, will often interpret laws in ways that the original authors didn’t anticipate. These interpretations can greatly extend a law’s scope and influence, such as the courts’ broad interpretation of what’s required to comply with NEPA’s “detailed statement” provision. Or they can reduce a law’s scope and influence, such as by determining that provisions of a law are unconstitutional. This happened with the 1989 Ethics Reform Act, which barred all federal government employees from being compensated for giving speeches, attending events, or writing articles. In 1995 the Supreme Court ruled that this requirement violated the First Amendment, and it only survives in application to senior government officials and Members of Congress.

Likewise, prosecutors or other government agencies might behave in ways other than what was expected. They might use a law for unexpected purposes: the DNA Fingerprinting Act of 2005, which authorizes collecting DNA from federal detainees, was part of a Department of Justice Authorization bill that was primarily focused on addressing violence against women. But the act was later used by ICE to collect DNA from immigration detainees. Alternatively, they might decline to use new legal machinery introduced. An example of the latter is the 2012 STOCK Act, which on paper made it illegal for members of Congress and their staff to trade stocks based on their congressional knowledge. As of 2025, there have been zero prosecutions under this law despite suggestive evidence that congressional insider trading does occur.

This category also includes Congress itself. Future Congresses may increase the effect of some law, such as by making a temporary program permanent or otherwise expanding its scope. For instance, the 1979 Recreational Boating Safety and Facilities Improvement Act contained a provision that created a trust fund, capped at $30 million fund, to clear a Forest Service tree replanting backlog. In 2021, the REPLANT Act took this fund and massively enlarged it, using it as a vehicle for a program to plant 1.2 billion trees in national forests. On the other hand, future Congresses might reduce the effect of some law: the 1976 Parole Reorganization Act, for instance, was intended to streamline and strengthen the federal parole system, but the 1984 Sentencing Reform Act abolished federal parole, making the previous law almost entirely moot.

You see the same sorts of divergences in the world at large. The 2006 Credit Rating Agency Reform Act tried to foster increased competition in the credit ratings agency market, but even though several new ratings agencies appeared, the market remained dominated by S&P, Moody’s, and Fitch, which collectively control 95% of the market. The Air Cargo Deregulation Act failed to predict how carriers like FedEx would respond to the freedom to fly on any route with any aircraft. TFTEA failed to predict how low-price Chinese fashion companies like Shein could take advantage of the “de minimis” change to ship directly to US consumers.

And of course, unforeseen behavior of legal operators and the broader world may interact. The Alaska Natural Gas Transportation Act failed to result in a new natural gas pipeline in part due to the market’s response to new natural gas regulations passed by Congress.

Conclusion

I think of laws as sort of akin to technology. With the invention of a new technology, you create some new capability, often for the purposes of achieving some particular goal. But once that capability is out there in the world, people will find all sorts of ways to take advantage of it. Marconi envisioned radio as literally “wireless telegraphy,” a way to send and receive messages from ships at sea, but he didn’t envision the rise of broadcast radio. Vacuum tubes were first used to amplify long-distance telephone signals, and only later became components for televisions and the earliest digital computers. Teflon was first used to make pump seals in uranium separation plants for the Manhattan Project, and only later found use in non-stick cookware.

Laws often work the same way. A law will create or modify some capability — an organization, a program, a rule that must be followed — aimed at accomplishing some particular thing. But once that capability is out there in the world, people might take advantage of it in different ways, finding uses for it that the creators of that capability never expected. A modest environmental reporting requirement becomes the foundation of modern environmental litigation; a minor change in employer retirement contributions becomes a retirement account used by tens of millions of Americans; a financial reporting law gets used to charge rioters.

Conversely, just because you introduce some new capability doesn’t mean it’ll actually be useful, or that anyone wants it. Some technologies, like 3D TV, or smell-o-vision, don’t pan out, and the patent archives are full of ideas for inventions that no one had any use for. Similarly, just because you create a new legal capability doesn’t mean it will end up useful in the way you envisioned. Changing the rules for designating a “nationally recognized” credit rating agency, as the 2006 Credit Rating Agency Reform Act did, did nothing to disrupt the market share of the existing agency oligopoly.

It is, of course, notoriously hard to predict the long-term effects of new technologies. With laws, it seems like predictions are substantially easier. But divergences still exist.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure, and industrial technology. This week we look at adults living with their parents, Samsung’s profits, Native American data centers, Puerto Rico’s electricity grid, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housekeeping items:

• No essay this week, but I’m working on a more involved piece about the predictability of legislation that should be out next week.

• The Origins of Efficiency was chosen as one of McKinsey’s 2026 Book Recommendations.

Housing and cities

The 21st Century Road to Housing Act becomes law after Trump declines to veto it. [NPR]

49% of adults under 30 in the US are now living with a parent, according to a recent Federal Reserve Survey. “...young people say that living at home in 2026 doesn’t carry the stigma it once did because of how unaffordable life has become. About 55% of young adults who moved back home said it was out of financial necessity, according to a spring survey by financial services firm Thrivent.” [WSJ]

Seattle has more empty downtown office space than any other major US city. [Seattle Times]

How the enormous success of TSMC has spurred development in the surrounding city of Hsinchu. “While the rest of Taiwan and much of East Asia are grappling with a sharp population decline, so many children have been born in Hsinchu in recent years that the schools cannot keep up. Public high schools in the area now admit only students with sufficiently high test scores, a practice that Ms. Lo and a group of mothers have protested. Those who do not make the cut must commute to another school district.” [NYT]

There was concern earlier this week that a New York highrise being renovated might collapse, after several columns began to severely buckle. Thankfully it looks like it’s been stabilized for now. “The massive renovation project at 219-235 E. 42nd Street — a conversion of the former Pfizer headquarters into about 1,600 apartments — was scrutinized Tuesday as construction workers had to be safely evacuated after the building began swaying. Damage to structural columns at the 21st floor is visible to the naked eye. Several nearby buildings were also evacuated as a precaution.” [CNN]

Last year New York City added more apartments — 38,682 units — than it has since 1965. This is still a pretty small number of units for a city New York’s size, to be honest: Austin added over 30,000 apartments in 2024. [WSJ]

CBRE released its 2026 look at US company headquarters relocations. California continues to lose headquarters, Texas continues to gain them. [CBRE]

Are rising insurance costs driving Americans to relocate from flood-prone areas? “High-flood-risk US counties — those classified in the top 10% for their share of homes that are very vulnerable to flooding — lost a net 63,357 domestic residents from mid-2024 to mid-2025, nearly double the outflow recorded over the previous 12-month period, according to an analysis released Wednesday by real estate brokerage Redfin. In 2024, flood-prone counties collectively posted a net population loss for the first time in five years.” [Bloomberg]

Manufacturing

Atomic Semi, the semiconductor startup famously started by Sam Zeloof after building microchips in his parents garage, rebrands as “Fab2.” “Fab2 designs and builds every tool in its fabs in-house, from pumps, valves, and gas lines to lithography and the vacuum chambers that house it. The company assembles those components into machines, the machines into complete fabs, and then aims to mass-produce the fabs themselves. It pairs the hardware with Studio, an in-browser, collaborative EDA tool for layout, schematic, and simulation work, previously branded as Atomic Studio.” [Tom’s Hardware]

The Economist on whether China got its hands on ASML’s EUV machines. “Impossible, says ASML. Europe’s most valuable company has told American officials that it knows the exact location of all 340 EUV machines it has produced, including 26 decommissioned ones. None is in China, it says. What is more, only ASML can transport the highly sensitive machines, which it monitors online, and components that it ships are handled by ASML engineers in customers’ fabs. “ASML has never shipped an EUV machine to China, nor have we shipped to China any component, module or equipment specially designed to be used in an EUV machine,” the company says.” [The Economist]

Thanks to the huge jump in the price of memory due to the AI boom, Samsung’s profits in its chip division for 2026 will be more than the previous 40 years combined. [Toms Hardware]

An interview with some of China’s first cohort of “practical PhDs,” graduate students who develop a product instead of writing a thesis. [Nature]

After losing more than $1 billion of them to Iran, the US wants a cheaper reaper drone. [Ars Technica]

Read more

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at households without homeowners insurance, crackdowns on AI chip smuggling, Japan’s two electrical frequencies, Meta’s AI compute business, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber!

Housing

Someone making the (somewhat dubious) point that fixed-rate mortgages have many similar impacts as rent control. The biggest difference, obviously, is that fixed-rate mortgages don’t disincentivize creating new housing supply the way that rent control does, but it’s interesting to see the similarities. “A homeowner with a 3 percent mortgage is, in practice, protected from the market in much the same way as a tenant in a rent-stabilized apartment. Both receive a valuable incumbent benefit. Both face a large penalty for moving. Both may remain in housing that no longer fits their needs because leaving would mean surrendering that benefit. And both systems impose costs on outsiders: prospective tenants in one case, prospective buyers in the other.” [Substack]

Apparently one in seven homeowners in the US don’t have homeowners insurance? [Insurance Dimes]

Manufacturing

It looks like we’re starting to see more crackdowns on smuggling AI chips into China from Taiwan. “Taiwan government agencies raided the offices of Super Micro Computer and several of its local affiliates, deepening an investigation into the alleged smuggling of Nvidia chips into China using the company’s servers.” [Japan Times]

Ford rehired a few hundred quality inspectors that it had tried to replace with AI. [BBC]

Sony will stop making physical discs for playstation games in 2028. [IGN]

The US Bureau of Industry and Security (BIS) denied Polestar, a brand spinoff of Volvo which is owned by Chinese company Geely, an authorization to sell cars in the US starting in 2027. Volvo, which is also owned by Geely, isn’t yet affected for some reason. “Polestar clearly didn’t see this situation coming. The automaker announced a reboot plan in February, which would’ve seen a slew of new product coming to the U.S. as the company grew the lineup. Global production of the Polestar 3 was moved from Chengdu, China, to Volvo’s Ridgeville, South Carolina, plant specifically to avoid the Trump Administration’s tariffs. The Polestar 3 currently rolls off the South Carolina assembly line alongside its platform mate, the Volvo EX90.” [The Drive]

China is building a factory to produce, among other things, personalized cancer vaccines. “The facility will house cell therapy research laboratories together with a production line of the company’s flagship product, LK101, a personalised cancer vaccine that analyses each patient’s tumour DNA to pinpoint the specific genetic mutations driving the disease. With AI, the company said the procedure could be completed in a day.” [SCMP]

Car companies are apparently switching to aluminum wiring due to the high costs of copper. [Reuters]

South Korea plans to spend $1 trillion on memory chip fabs and humanoid robots. “The most costly of the megaprojects involves a commitment by Samsung and SK Hynix of $585 billion to build new chip fabrication plants in the southwest provinces of South Korea, along with boosting semiconductor fab construction in the Seoul capital region, according to Reuters. The government’s goal is to double South Korea’s production of dynamic random-access memory (DRAM) within five years.” [Ars Technica]

Meanwhile, the Wall Street Journal on the delays to a Micron memory chip fab that was supposed to start construction two years ago, but has been held back by, among other things, bats. “Mr. Schumer has described Micron’s site as “open fields,” but it includes hundreds of acres of wetlands and forestland that are nesting areas for endangered bats. This makes permitting and building more complicated. Trees can only be chopped down when bats aren’t nesting—i.e., from November to March. To obtain federal and state permits, Micron committed to spend $1 million to protect the bats and install 10 “bat houses.” The manufacturer also agreed to provide on-site child-care for workers and enter into project labor agreements with unions in return for $6 billion in federal largesse.” [WSJ]

Read more

Every spring, as sure as the seasons, and for generations unknown, screwworms began their annual march northward from their overwintering sanctuaries in Mexico and South Texas. Pushed by an unknown force as inexorable as gravity, screwworms moved north — ever moving, ever spreading, ever multiplying, ever destroying. No army ever advanced any more surely or methodically. No army was ever more destructive. Attacking, killing, maiming, and destroying, screwworms literally ate their way north. Reaching upper South Texas, they fanned east and west - all the while moving north - dotting the countryside with the dead carcasses of hapless wildlife, cattle, sheep, and goats, filling the “wormy” pens of farmers and ranchers. - C.G. Scruggs, The Origin of the Screwworm Control Problem

Screwworms completely dominated our life. - T.A. Kincaid Jr., Texas rancher, quoted in The Peaceful Atom and the Deadly Fly

On June 3 of this year, a flesh-eating parasite, the screwworm, was found in a three-week-old calf near the Texas town of La Pryor. Since then, dozens more cases have been discovered in Texas and New Mexico. Outside of a screwworm outbreak in the Florida Keys in 2016 (which was contained), this marks the first screwworm infestation in the US since the 1980s.

Until now, the US has been free of screwworm not due to luck, but because of a decades-long program to eradicate the parasite by breeding it out of existence. By dropping millions of sterile male screwworm flies in an infested area, agricultural agencies can overwhelm the native, fertile male screwworms. Female screwworm flies, who only mate once in their life, will mostly mate with sterile males, producing no living offspring. Drop sterile flies for long enough, and eventually there will be no viable offspring at all, and the pest will be eliminated.

Over the course of several decades, this “sterile male technique” was used by the USDA to eliminate screwworm from the US, Mexico, and Central America. Since the early 2000s a joint US-Panamanian organization, COPEG, has maintained a “screwworm barrier” at the Darien Gap between Colombia and Panama. Every week, millions of sterile male screwworm flies are dropped over the gap, preventing the screwworms from spreading north from South America (where it remains endemic).

Sometime around 2023, the barrier at Panama failed, and for the last several years screwworm has been marching north. It’s now reached the US. Efforts are underway to eliminate screwworm from North and Central America once again, but it will likely be years before they succeed.

The screwworm eradication program was so effective at eliminating the parasite that we’ve collectively forgotten what an enormous problem it used to be. It’s worth understanding the costs inflicted by screwworm prior to its elimination, how a program emerged for eliminating it, and how control was allowed to lapse.

History of screwworm

New World Screwworm (scientific name Cochliomyia hominivorax) is a species of fly native to the Western Hemisphere. While the larvae (maggots) of most flies feed on dead or decaying tissue, screwworm is unique in that its larvae feed on living tissue. The grisly cycle begins when a screwworm fly lays its eggs on the open wound of an animal. The eggs soon hatch into wriggling white worms, which can grow up to two-thirds of an inch long. These worms burrow into the flesh as they eat their way into the animal, making the wound even worse and attracting even more flies to lay their eggs. After a few days of eating, the worms transform into shell-covered pupae, falling out of the animal and emerging as fully grown flies about a week later. Untreated, a screwworm infestation in an animal is typically fatal.

Since at least the early 19th century, and likely much earlier, screwworm was a miserable fact of life for raising livestock in the Southwestern US. The screwworm fly would find and lay her eggs on even the smallest open wound, and animals had to be inspected for worms constantly. One author notes that “[p]eople would not leave home for more than a day for fear of finding their animals had been eaten alive while they were away. Anyone who didn’t check their animals at least every two days — or have someone do it for them — knew that they would pay a heavy toll in damaged or dead animals.” When worms were inevitably found, they had to be dealt with, typically by applying various insect-killing chemicals, though occasionally by manually removing the worms, a task described as “disgusting and sickening.” Often animals couldn’t be saved — an infection in a horse, for instance, was described in the late 19th century as “typically fatal.” The amount of time and effort required was such that ranchers employed ranchhands whose entire jobs were inspecting animals for screwworm.

Despite efforts to prevent infection, such as by covering animals’ open wounds with tar, screwworm infestations occurred by the millions — in cattle, sheep, goats, horses, and occasionally humans (the first documented screwworm infection in a person in the US dates to the 1830s). One history of the scourge noted that by the early 20th century, infestations had reached “such a disastrous level that it was becoming unprofitable to raise livestock. Producers were forced to hire more and more cowboys to check and treat animals.” A 1935 USDA survey found more than 1.2 million infections and 180,000 dead livestock in Texas alone, with actual totals likely being far higher. A year earlier the USDA estimated that screwworm had killed 1.3 million animals across the Southeast. Some estimates suggested that in the 1930s and 40s, 60% to 80% of white tail deer in Texas were killed by screwworm infections.1

By the 1930s, the situation appeared grim. Screwworms had, by then, found their way into Florida, a climate warm enough that they could survive in the southern portion of the state year-round, spreading up through the southeast during the spring. Prospects for dealing with the pest appeared bleak:

Even with the best efforts of scientists and the use of the latest insecticides, repellents, and liver-baited traps, US livestock producers were losing the war. After more than one hundred years of struggle, the screwworm had spread out of control — and there was no solution in sight. - History of the Mexico-United States Screwworm Eradication Program

Studying the enemy

The toll inflicted by screwworm inspired vigorous efforts to try and combat it. In 1929 the USDA began a screwworm research program at Menard, Texas. This program initially focused on developing better chemical treatments for infections, and would ultimately generate the research that would banish the screwworm from the US.

The first breakthrough occurred thanks to USDA entomologist Emory Cushing. At the time the primary methods to battle screwworm were insecticides to treat infections and traps to capture the flies. After several years of research at Menard, Cushing became convinced that these measures wouldn’t be enough. Believing that something was missing from “the screwworm puzzle,” Cushing applied for a grant to study entomology at the University of Liverpool with Walter Scott Patton, a world expert on flies. Working under Patton, Cushing closely studied the internal structure of different flies, determining for the first time that the screwworm was a unique species with a unique lifecycle. Prior to this, the screwworm fly had often been confused with other, similar flies whose larvae fed on decaying, rather than living, flesh. The identification of screwworm as a distinct species, one whose larvae ate only living flesh, was critical: previous eradication efforts had included using traps baited with carrion to attract flies, but Cushing’s work showed that these efforts were useless. Following Cushing’s discovery, research efforts shifted focus to understanding the screwworm specifically.

In 1931, shortly before Cushing left for England another entomologist, Edward Knipling, joined the USDA. Initially posted at the Menard laboratory in Texas, Knipling was moved to a series of other labs, before eventually arriving at a newly formed lab in Valdosta, Georgia, established in 1934 to help combat the worsening screwworm problem in the Southeast. By then, thanks to Cushing’s research, it had become clear that the screwworm was a unique species, and Knipling and other USDA researchers began studying it more closely to understand and characterize its lifecycle and behavior. This meant watching the flies closely; at one point, Knipling spent every waking hour for a week straight doing nothing but watching screwworms all day. From “The Peaceful Atom and the Deadly Fly:”

He placed a wounded goat in the open. From daylight till dark for a week he watched the goat. Upon seeing a female screwworm fly deposit eggs on the wound in the goat, Knipling would mark her with fingernail polish. He would then watch the marked female and plotted all her activities for the rest of the week if she remained in the area. Day after day, fly after fly, Knipling observed. What time were flies most active? How many eggs were laid? Dozens of questions were asked. Knipling and his fellow workers tried to pry into the very basic life functions of the screwworm.

These studies revealed two important facts about the screwworm. One was that compared to other flies, the number of screwworms in a given area was surprisingly low: later studies eventually put the figure at around 100 flies per square mile. The other was that screwworms would not infect dead animals, only living ones.

In 1935 another young entomologist, Raymond Bushland, joined the USDA and was soon tasked with studying the effects of pesticides on screwworms. This required a steady supply of fresh screwworms, which could only be created by deliberately wounding caged animals allowing them to be infected. Bushland spent his days extracting screwworm pupae from the carcasses of dead rabbits, a task so repulsive that Bushland spent every morning of the first few weeks on the job vomiting.

Desperate to find a way to grow screwworms that didn’t require live animals, Bushland was, after months of work, eventually able to create an artificial growth medium consisting of “hamburger, blood, water, and a little formaldehyde to delay putrefaction.” The disgusting mixture made it much easier to grow screwworms, and before long Bushland was harvesting them by the thousands.

In 1937, Knipling, who had continued to move between various USDA locations, arrived at Bushland’s lab in Texas. Knipling was astounded at the huge number of flies that could be grown in the lab, compared to the relatively small number of flies that seemed to exist in the wild, and he began to ponder whether there might be some way to take advantage of the fact. Bushland’s screwworm husbandry operation also allowed Knipling to confirm a hunch he had from his own observations of the screwworm in the wild: the female screwworm fly only mated once in her life.

As Bushland and Knipling discussed their various observations and discoveries, the outlines of a plan began to emerge. What if you could overwhelm the wild population of screwworm by dumping huge numbers of sterile male flies into an area where cold temperatures had already reduced the number of wild flies? If sterile males greatly outnumbered wild males, most females would mate with sterile males, producing no viable offspring and greatly reducing the size of the next generation. Keep dumping sterile male flies, and eventually the population would simply breed itself out of existence.

It seemed like it would potentially work, but nothing like it had ever been tried before. And it wasn’t obvious how you could produce huge numbers of sterile male flies. After getting ridiculed by their entomologist colleagues — “who ever heard of castrating flies?” — Knipling and Bushland kept the idea to themselves. But the seed had been planted.

The plan to eliminate screwworm

For the next decade, a combination of professional skepticism, lack of obvious sterilization method, and the exigencies of World War 2 prevented much further work on screwworm eradication. At points during the Second World War Knipling continued to toy with models of screwworm population dynamics, and in 1947 he tasked Bushland with investigating chemical methods of sterilizing insects. But no method appeared promising. Without a way of mass-sterilizing the male flies, Knipling and Bushland’s idea was nothing but a pipe dream.

The final piece of the puzzle to eliminate screwworm wouldn’t fall into place until 1950. That year, Knipling read a journal article by Nobel Prize-winning scientist Hermann Muller. Muller had been awarded the prize for his groundbreaking work in genetics, discovering that mutations could be introduced into living things by exposing fruit flies to radiation. Muller’s 1950 article was a warning against the dangers of nuclear war, cautioning that the resulting radioactive fallout could create “a world of sterile human beings.”2 Knipling contacted Muller and asked him if radiation could be used to create large numbers of sterile male screwworms. Muller responded: “I know nothing of screwworms but your theory is sound.”

Invigorated by the Nobel Prize-winner’s vote of confidence, Knipling and Bushland set to work to see if they could use radiation to sterilize screwworms. With no research funds, Bushland dedicated his nights and weekends to the project, finding creative ways to make progress. Muller had done his radiation research using X-rays, but when a local hospital wanted $200 per test to use an X-ray machine, Bushland leveraged his Army connections to use the X-ray machine at a nearby Army hospital for free. The results were promising. Wanting to try an even more powerful radiation source, Bushland was able to convince Oak Ridge National Lab to lend him a sample of highly radioactive cobalt-60, capable of generating gamma rays. Meanwhile, Knipling was eventually able to secure $20,000 in research funds from the Atomic Energy Commission, who were interested in radiation’s effect on fly sterility.

The tests were a success: with a high enough dose of radiation, males were completely sterilized. They survived and would still mate with females, and the females would lay eggs, but the male chromosomes had become so riddled with mutations that the eggs didn’t hatch.

With a method of industrialized screwworm sterilization in hand, the only thing left to do was to try the idea and see if it worked. A USDA team in Florida began working out ways to distribute huge numbers of sterile screwworm flies.

They first tried simply scattering sterilized screwworm pupae on the ground, but this didn’t work: ants and other insects quickly devoured them. Dropping live flies seemed like it would work better, but it would have to be done in an isolated location to see if elimination was really possible — otherwise the area would be continuously reinfested by screwworms from outside the test area. After looking at several possible locations, the team eventually settled on Sanibel Island, a small island near Fort Myers on Florida’s Gulf Coast. After a few months of dropping boxes of flies onto the island from a small plane, the number of wild screwworms dropped to virtually nothing. While they couldn’t be eliminated completely — likely because wild flies continued to migrate from the mainland — the results were highly encouraging.

Buoyed by their success, the team began looking for an even larger stage for their next test. By chance, as they were searching, Knipling received a letter from a Dutch agricultural officer recently stationed on the Dutch Caribbean island of Curaçao, off the coast of Venezuela. The island was overrun with screwworm, and the officer wondered if Knipling had any suggestions for dealing with it. The location seemed perfect, and Curaçao soon became the location of the next test.

The Curaçao test began in the summer of 1954. A screwworm production facility, complete with a cobalt-60 gamma-ray source, was set up in Orlando. Sterilized screwworms were flown into Curaçao, then dropped from a small plane in a carefully planned pattern across the island. For the first few weeks, results were poor, with 85% of screwworm offspring still viable. But when the density of dropped sterile flies was raised from 100 to 400 flies per square mile, results quickly improved. Within 14 weeks, no viable offspring could be detected. In November 1954, Curaçao was declared free of screwworm.

Pushing the enemy back

With the success in Curaçao, the USDA was ready to try eliminating screwworms in the US mainland. In 1955 a new, larger screwworm factory was constructed on the outskirts of Orlando, capable of producing two million sterilized flies a week. This new insectary would use a new strain of fly, bred specifically to outcompete wild flies. (The new flies, which had been selected for their high mating frequency, were nicknamed “the sexual athletes”.) Thanks to lobbying from the Florida Cattlemen’s Association, whose members were collectively losing $20 million per year due to screwworm infestations, the Florida legislature allocated $3 million to the eradication program, a figure which was matched by the federal government.

In April of 1958, while screwworms were still confined to the southern half of the state due to an unusually cold winter, USDA planes began dropping sterile screwworm flies over Florida. Flies were first dropped across a 100-mile-wide swath cutting through the center of the state, effectively creating a “barrier” that flies in the southern half of the state wouldn’t be able to penetrate. Inspection stations were set up to prevent the movement of any infected livestock into northern Florida and Georgia, and by July no screwworm cases were being reported north of the quarantine line. The USDA then pushed into south Florida, aided by an even larger screwworm factory in the city of Sebring capable of producing 50 million flies a week, and by September reported screwworm infections in the southeast had dropped to virtually nothing. Over the next several months, the occasional outbreak occurred, but these were dealt with by carpet-bombing the area with ever-higher concentrations of sterile flies. By February of 1959 screwworm cases in the Southeast had dropped to zero.

By the early 1960s screwworms were inflicting more than $100 million worth of damage in the southwest annually. Impressed by the success of the Florida program, ranchers in Texas and the Southwest began clamoring for similar efforts. But Texas would prove to be a much tougher nut to crack than Florida. Unlike Florida, which was surrounded on three sides by water, Texas had no natural barriers preventing screwworm transmission across the 1,000+ mile border that the state shared with Mexico, where screwworms survived year-round. In October of 1959, the USDA issued an official statement that screwworm eradication in the US Southwest did not appear feasible.

But despite the USDA’s official stance, Knipling, Bushland, and a few others in the USDA realized that a strategy used in the Florida eradication program might work in Texas and the Southwest. Florida’s eradication effort began by creating a 100-mile-wide screwworm barrier across the center of the state. This barrier, once in place, had virtually halted the spread of screwworms into the northern half of the state. What if the same thing were done in the Southwest? Create a 100-mile-wide barrier across the entire US-Mexican border. The challenges were many: this barrier would stretch far longer than the one that had been created in Florida and, unlike in Florida, this barrier would have to be continuously maintained. It would also require eliminating screwworm everywhere north of the barrier, a potentially massive undertaking. But not only could this plan eliminate screwworm from the Southwestern US, it could prevent re-infestation of the Southeast, a constant threat that was warded off by an involved, expensive livestock inspection operation.

For the next several years, the official position of the USDA remained that a Southwestern elimination program wasn’t feasible. But eventually the USDA changed its policy, thanks to the efforts of the Southwest Animal Health Research Foundation (SWAHRF), an organization formed by a small group of Texas livestock producers. SWAHRF broke the logjam by raising millions of dollars in voluntary donations from Texas ranchers for screwworm eradication. The show of grassroots enthusiasm convinced the Texas Legislature to allocate money to a Southwest eradication program, which in turn helped convince the federal government to support it. Credit also goes to newly elected Vice President Lyndon Johnson, who owned a ranch, was familiar with the scourge of screwworm, and pulled strings to kickstart the program.

The Southwest eradication program began in February of 1962, when winter temperatures had pushed the screwworm down into Mexico. USDA aircraft began dropping sterile flies across the US-Mexico border to create the North-South barrier. But the effort failed, and before long Texas had been reinfested. There simply weren’t enough flies: sterilized flies were being produced at a small insectary near Kerrville, Texas, at a rate of 20 million per week, but creating an effective barrier would require more than ten times that rate of production.

By spring 1962, construction had begun on the biggest screwworm factory yet, at an abandoned Air Force base near Mission, Texas. Eventually capable of producing more than 200 million screwworm flies a week, the Mission factory was a grotesque marvel of insect-producing efficiency. Operating 24 hours a day, 365 days a year, it was, in essence, a 76,000-square-foot artificial wound. Trays full of meat, blood, and water, each one heated to the exact right temperature to stimulate screwworm growth, moved through the facility on a monorail system timed to the lifecycle of the screwworm. Eggs would be placed in on the trays, hatch into larvae, and collectively feed on the bloody nutrient sludge, creating a “seething mass that is difficult to believe unless you’ve seen and smelled it.” After several days of growth, the larvae would wriggle out of the feed and fall into a water-filled trough, which gently carried them into sawdust-filled trays where they would pupate. The pupae would then be collected into canisters, which in turn were loaded into large casks containing highly radioactive cobalt-60. The irradiated, sterilized pupae were then packed into cartons, loaded into refrigerated trucks, and sent to distribution centers, where they would eventually hatch, be loaded onto planes, and dropped by the millions.

By June of 1962 the Mission factory was fully online, and flies began to be dropped in July. However, the program soon made an unfortunate discovery. It had previously been believed that the maximum range of the screwworm fly was around 35 miles traveled per week. Because the fly only lived for a few weeks, a 100-mile barrier was believed to be sufficient to prevent the passage of migrating flies. However, recent research had shown that in some cases a screwworm could fly much farther — up to 180 miles in a week. Consequently, the barrier would have to be much wider as well: 400 miles. Left with no choice, the USDA ramped up fly production at the Mission factory as fast as possible to support a widened barrier, and an agreement with Mexico was worked out to allow USDA planes to fly hundreds of miles into Mexican airspace to drop sterilized screwworm flies.

The expanded barrier worked. By 1963, the rate of screwworm cases in the US had fallen by 90%. In 1964, they fell even more, and by 1965 the states of Texas, New Mexico, Oklahoma, and Louisiana were declared screwworm free. California and Arizona soon followed, and by 1966 screwworm had been eliminated from the entire US.

From Mexico to Panama

But the national elimination of screwworm from the mainland United States was hard-won and not easy to maintain. It required continuously dropping millions of sterile male flies across the US-Mexico border to create a screwworm barrier. And no barrier is perfect. Over the next several years screwworm infections continued to crop up in the US, as worms from Mexico found their way into the US. 1972 was particularly bad, when US cases (which had fallen as low as 170 in 1970) surged to more than 95,000 that year.

The solution? Push the US barrier even farther south, down to the Isthmus of Tehuantepec in southern Mexico. Not only would this reduce the likelihood of Mexican screwworms reaching the US, it would eliminate screwworm from most of Mexico. And a barrier at Tehuantepec, which was only 118 miles across at its narrowest point, would be far easier to maintain than the nearly 2000 mile barrier across the entire US-Mexican border.

In 1972, the US and Mexico signed a Screwworm Eradication Agreement, creating a joint commission to oversee the elimination of screwworm from northern Mexico. A new screwworm insectary, which would eventually produce more than 400 million sterilized flies per week, was built near Tuxtla Gutiérrez in southern Mexico. The factory began producing flies in 1976, and the first flies were dropped on Baja, Mexico that September. Progress was initially slow, but after a program reorganization (which included more livestock inspectors and a new strategy for quickly deploying experts to areas where outbreaks occurred) results improved. By late 1979 the Baja peninsula was free of screwworms, and by 1980 many Northern Mexican states followed. Starting in 1981, the program established a “critical line” across the width of Mexico, marshalling all possible resources to squash any infections that appeared north of it. Over the next several years the critical line was steadily pushed downward, until it reached the Isthmus of Tehuantepec in 1984, one year ahead of schedule.

But establishing a final barrier across the Isthmus of Tehuantepec left something to be desired. Mexican livestock producers below the barrier, who continued to deal with screwworm, complained that the government was favoring northern producers, and hundreds of thousands of livestock that passed from the south to livestock markets in the north, requiring tedious inspection and creating a constant risk of reinfestation. And while the Tehuantepec barrier was much smaller than the US-Mexico one, it still required dropping 150 million sterile flies a week, every week.

Fortunately, an even better location for a barrier existed: the Darien Gap, on the border of Colombia and Panama. At this narrow stretch of land, the barrier would need to be just 60 miles wide. And the densely forested, difficult-to-navigate terrain, which had no roads crossing it, would further hinder the passage of any infected livestock. Pushing the barrier all the way to Panama would also have the benefit of eliminating screwworm not just from the rest of Mexico, but all of Central America as well.

Eradication in Central America posed new challenges: the favorable climate conditions for the screwworm meant that far more flies would need to be dropped per square mile, and new international agreements would need to be signed with each country involved to allow the USDA to fly over their airspace and drop millions of insects. But the benefits would be substantial, and the decision was made to extend the program.

In 1986 the US-Mexico commission signed an agreement with Guatemala, and began dropping flies over the country in 1988. That same year an agreement with Belize was signed, which was followed by agreements with El Salvador, Honduras, and Nicaragua in 1991, Costa Rica in 1993, and Panama in 1994. As sterile fly dispersal began across the new territories, screwworm infections in each country quickly fell. In 2006, a new screwworm factory was built in Panama to replace the aging Mexican factory (which had been operating continuously since 1976), opening on the same day that the country was declared screwworm free.

For the next decade and a half, the barrier at Panama held. Administered by a joint US-Panamanian organization called COPEG, the program continued to drop millions of flies a week across the Darien Gap, and employed inspectors to monitor livestock in the surrounding area. When the occasional infestation flared up — in Aruba in 2011, the Florida Keys in 2016 — flies kept in reserve were quickly dispatched, dropped in huge numbers to quash any spread.

The barrier falls

At some point in the last few years, screwworms broke through this barrier.

In 2021, screwworm cases were confirmed within the Darien Gap area in Panama, and Panama was added to the list of countries where screwworm is confirmed by the USDA’s Animal and Plant Health Inspection Service (APHIS). Initially, this didn’t appear concerning: since 2001, Panama has had occasional small screwworm outbreaks in the Darien Gap area. But in 2022, screwworm was detected in Panama beyond the barrier at the Darien Gap. Though initially this was just a few cases, the infestation quickly spread: in 2023 there were 6,500 screwworm cases in Panama, and cases were detected in nearby Costa Rica. By 2024 screwworm had spread to every Central American country, as well as Mexico; there were 18,000 detected cases in Panama, 8,600 in Costa Rica, and 3,300 in Nicaragua. In 2025, total cases in Mexico rose to over 12,000, and as of this writing have exceeded 30,000.

It’s not 100% clear what caused the breach, and most sources point to a confluence of different factors. The disruption caused by COVID-19 seems to be partly to blame: during the pandemic livestock inspectors were forced to stay home, vehicles broke down and couldn’t be repaired due to a lack of replacement parts, and power outages in the screwworm plant killed millions of sterilized flies.

Another issue seems to be large-scale movement of both people and livestock. Starting in 2021, the number of migrants passing through the Darien Gap skyrocketed. Illegal cattle trafficking also likely played a role. Ranchers throughout Central America will illegally graze cattle on protected forests, often as part of a larger money-laundering operation for narcotics cartels. These cattle are then illicitly moved north to livestock markets in Mexico. It’s been estimated that hundreds of thousands of cattle are trafficked through Central America each year, and the speed at which screwworms have spread through the region suggests they’re moving by truck, being carried by infected cattle.

Also contributing are the changing circumstances in the Darien Gap itself. When the barrier was established the Darien Gap was a wild, untamed rainforest. Today that forest is increasingly being cut down and replaced with grassland for cattle to graze on, making it much easier for a screwworm infection to spread through the area. Amplifying this risk is the fact that many ranchers are absentee owners, people who treat the ranch as something of a vacation property without closely monitoring its operations.

The USDA didn’t simply sit back in response to this spate of outbreaks in Central America. APHIS drew on hundreds of millions of dollars in emergency funding in 2023 and 2024, and began dropping sterile flies in infected countries. Cattle imports from infected countries ceased, (though they later reopened in a limited fashion with increased inspection protocols). Fly production at the Panama facility was ramped up from 20 million flies a week to more than 100 million. But these efforts were nowhere near enough to stop the spread. The facility at Panama could produce enough flies to maintain a dense, narrow barrier of sterilized flies at the Darien Gap, but not to fight an infestation which had spread widely throughout Central America; at the peak of Central American operations in the 1980s and 1990s, over 400 million flies were being produced weekly.

New screwworm facilities are coming online — in 2026 the USDA broke ground on a facility at Moore Air Force Base in Texas meant to produce 300 million flies a week, and is converting an existing fruit fly insectary in Mexico to produce screwworms. And earlier this year APHIS announced a $100 million “Grand Challenge” for projects related to screwworm eradication or treatment. (Some anti-screwworm efforts may have been hindered by DOGE, which cut APHIS staff, screwworm monitoring programs, and may have delayed funding for the Mexico facility, but it’s hard to be confident about this, and the administration has unsurprisingly rejected these claims.)

Overall, the reemergence of screwworm seems like it’s being taken quite seriously, and if this level of investment is maintained it seems likely we’ll be able to re-eradicate it from the US and Central America. But this will probably take “close to a decade of sustained work:” bringing new production facilities online, and dropping millions of sterile flies over the infected areas week after week, month after month, year after year as the pest is gradually pushed back.

Conclusion

Overall, the screwworm program seems like a classic case of something becoming a victim of its own success: a problem got solved so thoroughly that we forget how big of a problem it was, and we gradually undermine the conditions that made the solution possible. Prior to the push of screwworms all the way south to Panama there were a series of other screwworm eliminations that only partly solved the problem: eliminating it in Florida and the Southeast didn’t do anything for the livestock in the Southwest, and required an involved inspection regime to keep screwworm entering from the rest of the country. Pushing the pest population down to Mexico required continuously dropping hundreds of millions of flies over an enormous, 400-mile-wide barrier that was frequently breached. The barrier at Tehuantepec left much of Mexico still exposed to screwworm, and was expensive to maintain.

The barrier at Panama, by contrast, worked amazingly well and cost only $15 million per year to operate. As a result of this success, other screwworm facilities were shut down (the Tuxtla factory closed in 2012, against the protests of COPEG), and production at Panama gradually declined to the level needed to maintain the barrier and not much else. Barrier conditions in the Darien Gap were allowed to deteriorate, the dense rainforest replaced with grassland occupied by ranchers paying insufficient attention to their herds. What had once been very thorough cattle inspection processes were allowed to lapse: one article notes that in Panama ranchers, veterinarians, and trade organizations became less diligent about inspecting cattle prior to shipping them, and that “a carbon copy of all these issues happened in every country all the way to Mexico.” And when it was clear that screwworms had breached the barrier, responses were sometimes delayed by political disputes — Mexico apparently initially made it very difficult for USDA screwworm flights to operate until the US Agricultural Secretary called to force the issue.

Unfortunately, it looks like we’re going to re-learn how much of a problem screwworm was the hard way.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at Trump refusing to sign a housing bill, the high cost of US-made doors, slow trucking, why we stopped making new land, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housing

The ROAD to Housing Act has now overwhelmingly passed both chambers of congress (358-32 in the House and 85-5 in the Senate). [NBC News] But in a last minute twist, Trump has said that he won’t sign the bill until another piece of legislation, the SAVE America Act, passes first. “Hours after his own aides praised the bill and promised the president would sign it, Mr. Trump instead canceled a scheduled event at the Capitol. Eschewing an opportunity for rare bipartisan accord, the president opted to turn the bill into political leverage, aiming to force Congress — and members of his own party — to bow to unrelated demands over voting restrictions and the war with Iran.” [NYT] Not exactly clear how this will play out, since (theoretically) there’s sufficient votes to overcome a presidential veto.

Claims from the WSJ that the cost of homeownership is rising faster than overall inflation, though the sourcing here seems a little thin to me. [WSJ]

The National Institute of Standards and Technology (NIST) has released technical findings on the causes of the collapse of a condo building in Surfside, Florida in 2021. Apparently the failure was in large part due to failing to conform to building codes when the structure was built (due to both design and construction errors), which carved away what should have been a substantial safety margin in the loading applied to the building. [NIST]

One issue I worry about with stagnant construction productivity is that without being able to drive down costs by way of technical improvements, builders in a competitive market might be more incentivized to cut costs at the expense of quality. We’ve previously noted that the largest US homebuilders seem to have a rapidly rising rate of construction defects, and that trend seems to be continuing. “D.R. Horton and Lennar, the two biggest builders in the U.S. by total volume, have experienced the biggest surges in potential legal costs. Lennar’s self-insurance reserve, earmarked for liabilities that insurance won’t cover, rose 21% in fiscal 2025 to $336.9 million, according to the company’s annual financial statement. D.R. Horton’s reserves for legal claims, which include expectations for future claims, rose 57% to $1.1 billion from the end of fiscal 2022 to the end of fiscal 2025.” [WSJ]

A provision in the 2021 bipartisan infrastructure package requires that any infrastructure funded with federal dollars use US-produced construction materials. “Infrastructure” includes homebuilding, so any homes built or funded by the Department of Housing and Urban Development are hit with this requirement, which can dramatically drive up the construction cost of these homes. “Outfitting a 40-unit building with standard imported door sets runs about $67,000. The only American-made equivalent is a hospital-grade product built to keep germs off surfaces in an operating room, and it is priced like one: $546,000. More than eight times the cost, to open and close the same doors…On the Davenport job, the only compliant elevator the team could find cost about $75,000 more than the standard unit. The switchgear and the HVAC told the same story: the American product existed, but only at a premium.” [Substack]

Works in Progress on how smashing NIMBYs created modern capitalism. [Works in Progress] And Works in Progress on how Europe’s housing problems are worse than America’s. [Works in Progress]

Manufacturing

The Mexican government wants to build a locally produced, inexpensive EV. [Gizmodo]

An open letter against California’s AB2047, which requires 3D printers to have detection algorithms to prevent the manufacture of firearms. “AB2047 requires every 3D printer sold in California to run a DOJ state-certified “detection algorithm” - a technology that can not reliably exist. If passed, it would pull a tool used in thousands of schools, libraries, labs, and small businesses out from under our communities.” [3D Printing Nerd]

US EV manufacturer Lucid lays off 18% of its workforce, right on the heels of a 12% layoff earlier this year. [CNBC]

On the other end of the spectrum, Chinese EV sales in Europe seem to be surging. “Leapmotor’s sales surged ​465.1% ⁠in May, while Chery and BYD jumped 244.1% and 136.6%, respectively. Among other manufacturers, Geely and SAIC recorded increases of 12.6% and 13.9%.” [Reuters]

How Slate got the price of its EV truck down to $25,000. [Heatmap] Others have pointed out that despite the hype, this isn’t that much cheaper than an entry level Ford Maverick. [Ford]

Read more

Summary

US subway construction follows the NFPA 130 fire safety standard, which mandates cross-passages — side tunnels that connect adjacent train tunnels — every 800 feet. This spacing is significantly closer together than European cross-passage spacing requirements, adding to construction costs without measurably improving safety. NFPA 130 requirements should be changed to align more closely with European cross-passage spacing requirements, or agencies and jurisdictions should modify these standards themselves when adopting NFPA 130 or requiring compliance.

Problem

The construction of anything, be it a new home, an office building, or a subway tunnel, is governed by a series of codes and standards that dictate how it should be designed and built. These codes specify performance requirements (e.g., a residential floor must support a certain amount of weight) and analysis methods (e.g., the capacity of a concrete beam should be calculated using a specific formula), and determine which materials or components must be used (e.g., buildings of certain heights must be made of steel or concrete). These requirements vary across countries — or even regions — depending on the codes and standards adopted in a particular jurisdiction. Often, these requirements are intended to keep occupants or users of a piece of infrastructure safe and ensure that a building won’t collapse during an earthquake, be blown away in a hurricane, or burn rapidly in a fire.

Subway systems, like other parts of the built environment, are subject to various codes and safety standards. These systems often consist of two or more adjacent tunnels to allow trains to travel in different directions. One code requirement stipulates the maximum distance between cross-passages, the smaller tunnels that connect the larger tunnels together to allow egress from one to another. This is a safety consideration: if there’s a fire in one tunnel, cross-passages allow riders to take refuge by providing access to an adjacent tunnel. Cross-passages are often equipped with fireproof doors to prevent the spread of fire and smoke.

US subway systems typically follow the cross-passage spacing requirements dictated by the National Fire Protection Association’s “Standard for Fixed Guideway Transit and Passenger Rail Systems,” more commonly known as NFPA 130. The National Fire Protection Association is a nonprofit organization made up of volunteer fire safety experts that publishes over 250 codes and standards on various aspects of fire safety. Because the NFPA code (like most US building codes and standards) is authored by a private organization, it doesn’t hold the force of law and is only required when adopted or mandated by a transit agency or jurisdiction (New York and Washington state, for instance, have incorporated NFPA 130 into their respective building codes). If a mass transit project receives federal funding — as most transit projects do — the Federal Transit Administration (FTA) requires the project to conduct a safety certification, which typically involves complying with NFPA 130.

Code requirements can often add costs to a construction project, as they require additional systems, stronger materials, and additional inspections. In the US, there is famously a large jump in building construction costs when going from seven to eight stories, because the marginal story triggers stricter building code requirements. Subway cross-passage requirements are no different. While the primary tunnel for a rail system can be efficiently dug using a tunnel-boring machine, cross-passage construction must be done using other, less efficient methods, such as drill-and-blast (where holes are drilled into the rock, filled with explosives, then detonated) or using a roadheader (a tunnel construction machine with a large boom-mounted cutting head). Cross-passage construction has been described as “one of the most challenging construction stages for any twin-tube construction project.” The UK’s HS2 high-speed rail project suggests that each cross-passage might cost on the order of $1.2 million to construct, and a 2019 geotechnical report from a Dallas Light Rail project states that cross-passage construction costs range from $100,000 (for short passages in sound rock) to more than $2 million (for longer passages in complex conditions); a 2022 report by the Miami-Dade Transportation Planning Organization on tunneling technologies for a Miami transit system estimated that cross-passages would cost $5–10 million each. The more cross-passages code requires, the more expensive a subway system will be to construct.

Because code requirements add cost, it’s important to understand their value, and to what extent they actually increase safety. With subway cross-passages, it’s notable that NFPA 130 spacing requirements used in the US are significantly tighter than European requirements. NFPA 130 requires that cross-passages in rail tunnels be spaced not more than 800 feet (244 meters) apart.

More specifically, they must be at least 800 feet apart when the distance between stations exceeds 2,440 feet. This spacing requirement originated in the design for Atlanta’s MARTA metro system in 1970: 800 feet was the distance it was believed passengers could walk downstream of a fire before flashover, when all combustible materials in an enclosed area ignite. However, EU standards require a maximum spacing of 1,640 feet, more than double NFPA’s required distance. The London Railway’s Elizabeth line, completed in 2022, has cross-passages spaced at roughly 1,600 feet, and in some places spacing between passages reaches up to 2,275 feet. Many long, recently built European tunnels have used a cross-passage spacing closer to the 1,640-foot limit, but still larger than the 800-foot maximum required by NFPA 130. Moreover, a 2010 analysis of metro systems around the world found that cross-passages were uncommon in most existing European and Japanese metros.

There’s little evidence that the tighter cross-passage spacing of NFPA 130 improves subway fire safety. Both European and US subway systems are exceptionally safe: the National Safety Council notes that the rate of passenger fatalities on urban rail transit in the US was about 15–50 times lower than driving a car, and rail transport has similarly low fatality rates in Europe.

London and New York have similar metro systems, each roughly 250 route-miles and transporting roughly 1–1.5 billion riders annually. London has had two major casualty subway events in the past 40 years — the King’s Cross Station fire in 1987, and the London Underground terrorist bombing attacks of 2005 — neither of which would have been ameliorated by tighter cross-passage spacing.1 Excluding those two events, both London and New York have had vanishingly few passenger deaths: seven and eight, respectively, in the last 30 years.2

The NFPA has admitted to a “lack of technical substantiation” for its cross-passage spacing requirements. A Japanese rail consultant analysis of NFPA 130 cross-passage requirements, performed as part of the planning for Cairo’s new Metro Line 4, noted that passenger trains don’t carry flammable materials and are generally made of noncombustible materials, and there is high probability that a train can make it to a station in the case of fire. The analysis recommended that cross-passages not be added on Metro Line 4, as they would “raise the cost and extend the construction period but will not enhance the safety so much.” A 2019 presentation by metro rail consultant Mosen similarly noted that the cost-benefit ratio for additional cross-passages is often estimated to be much greater than 10 (that is, costs are more than 10 times as high as benefits).

NFPA 130 does currently allow greater cross-passage spacing if an engineering analysis demonstrates that it will result in an acceptable level of safety. However, this places a significant burden on the transit project design team, as it requires them to perform an analysis and risk failing to secure approval, possibly delaying the project. Thus even with this escape hatch, 800-foot spacing remains the de facto requirement, and tunneled transit construction costs are higher as a result.

Solution

Due to the high cost of constructing additional cross-passages and their questionable safety benefits, the NFPA should update its cross-passage safety requirements.

NFPA 130, like most building and infrastructure standards, is periodically revised. The code was most recently updated for 2026, with the next revision scheduled for 2029. The NFPA accepts public comments and suggestions for code changes; comments on the first draft of the 2029 edition closed in April 2026, but are open until June 2, 2027 for the second draft. Federal regulators and other experts should recommend that the code’s cross-passage spacing requirement be updated to match European standards, and NFPA officials should strongly consider revising these requirements.

To build support for these recommendations, federal regulators could perform a formal study comparing US and international cross-passage requirements, similar to the 2022 study led by the Federal Railroad Administration (though such a study would have to begin soon to meet the second draft comment deadline).

Because the NFPA is a private organization whose standards don’t, by default, carry the force of law, transit agencies and local jurisdictions could also simply modify cross-passage spacing requirements directly. To the extent that FTA requires compliance with NFPA 130 to achieve a safety certification, it could update its policies and allow cross-passage spacing of up to 1,600 feet (~500 meters). State and local jurisdictions and agencies that have adopted NFPA 130 could similarly adopt modified versions that include more reasonable cross-passage spacing requirements. This sort of modification is widely practiced by jurisdictions that adopt other model codes, and is already exercised by some jurisdictions following NFPA 130: both New York and Washington state made modifications to NFPA 130 before adopting it. The previously mentioned study comparing US and international cross-passage requirements could bolster this effort as well.

Our knowledge of subway safety and performance has grown over time. NFPA 130 cross-passage spacing requirements were originally based on what was believed would enhance subway safety in the case of a fire. However, we now have several decades of evidence that these safety benefits are illusory, while the costs are substantial. The standards should be reconsidered.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure, and industrial technology. This week we look at a new housing bill, General Motors joining the grid-scale battery game, skepticism about data center delays, solid-state air conditioning, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housekeeping items:

• This week IFP published its Transit Abundance Playbook, a collection of 15 specific policy ideas for bringing down the cost and time it takes to build transit in the US. I have a piece in the playbook here.

Housing

The US house and senate have reached a deal on a combined version of the Senate’s ROAD to housing act. The combined version looks like it’ll be a good bill, as the burdensome build to rent provisions have been eliminated. [X] [Politico]

Do housing developments that include grocery stores like Costco or Safeway have a better chance of succeeding? “The grocery anchor brings a constituency that ordinary apartment developments lack. A developer fighting NIMBYs alone is on hostile rhetorical terrain. The actual political economy is more complicated, but at community-meeting altitude “developer” is shorthand for “rich greedy person,” and that’s the altitude these fights are conducted at. A developer fighting NIMBYs with Costco or Safeway as co-sponsor is in an entirely different argument. Costco is consistently rated one of the most-trusted retailers in America. Its CEO Ron Vachris has described the Coliseum project as a way “to enter markets where traditional big-box development would be nearly impossible,” and the Coliseum alone is expected to create 400 permanent retail jobs plus thousands of construction jobs. The Marina Safeway has the kind of intergenerational neighbourhood loyalty no developer can manufacture by himself; opposing 800 flats above it is one thing, but opposing the expansion of the Singles Safeway (57% bigger, with a wider deli counter and more checkouts) is much stranger to argue at a community meeting.” [Governance.fyi]

Works in Progress on how the Squamish Nation in Canada went about building the enormous “Senakw” housing project in Vancouver. “Senakw has an unusual history. The land it is built on was home to the Squamish people until they were forced out in 1913. Almost a century later, a court case restored the land to the descendants of those who were expelled, along with almost 100 million Canadian dollars in compensation. Freed from the restrictive planning rules that hold back densification in the rest of Vancouver, the Squamish decided in 2019 to use the land to build apartment blocks that, as well as housing Squamish people, are expected to generate around C$10 billion in income, equivalent to more than two million per person.” [Works in Progress]

Manufacturing

Semiconductor startup Phoenix Semiconductor is building replacements for the obsolete, no-longer produced microchips used by long service-life equipment such as military jets. “When the U.S. Navy’s fleet of F/A-18F Super Hornet fighter jets is headed for the scrap heap because an essential chip is unavailable, what do they do? They turn to Ryan Hatcher, the CEO and founder of Phoenix Semiconductor. Hatcher repackages off-the-shelf semiconductors into devices that are virtually identical to the phased-out chips.” [IEEE Spectrum]

Also on the subject of military manufacturing, the Trump Administration has invoked the Defense Production Act to deal with supply constraints in munitions manufacturing. [Reuters]

Indium phosphide is a mineral used for light-detecting semiconductor devices, such as fiber optics. China controls 70% of the world’s supply of indium, and issued export restrictions on indium phosphide in 2025. Those restrictions are starting to bite. “The U.S. urgency to resolve China’s export controls on the compound highlights how indium phosphide (InP) has emerged as a powerful trade weapon for Beijing that experts and executives say could disrupt the global rollout of AI data centres. “InP is one of several supply chain bottlenecks collectively gating AI data centre buildouts,” said Konrad Wang, a research analyst at SemiAnalysis. With AI workloads ​growing exponentially, InP is in high demand as it is a core material with no substitute in the new technology that data centre developers are turning to - using light through optical fibres, or photonics, instead of electrical signals ​through copper wire.” [Reuters]

Not satisfied with the tariffs on Chinese EVs, two Michigan lawmakers want to ban Chinese cars from even entering the country. [Techdirt]

The IEA released its 2026 electric vehicle trends report. [IEA] And RAND released a long report on how China’s industrial policy has evolved over the last 10 years or so. [RAND]

The US is worried China might have one of ASML’s most advanced EUV lithography machines. [Bloomberg]

Read more

The rise of the gas-powered automobile in the early 20th century shifted a substantial portion of coal consumption to petroleum, but coal still remained favored for industrial power. And this didn’t change with the emergence of the electric power grid: Thomas Edison’s first central electricity generating station at Pearl Street in New York used coal-fired reciprocating engines, and coal was the primary method of generating electric power in the US well into the 21st century.

By the end of the 20th century, however, this trend was starting to shift. For most of the 20th century coal made up around 50% of US electricity generation, but after peaking at around 57% of electricity generation in the mid-1980s, coal started to decline as a share of electricity generation in the US. And starting around 2008, coal-generated electricity began to decline in absolute terms, falling from over 1.6 trillion kilowatt-hours produced in 2009 to around 0.8 trillion in 2020. Today, coal supplies around 16% of US electricity, a share that seems likely to continue to fall long-term.

As coal became less popular, many coal plants — over 200 since 2008 — have simply shut down.1 But some of these plants were instead converted to burn natural gas in place of coal. Since 2008 there have been around 140 such conversions.

Given recent attempts to reinvigorate the coal industry, with the Trump Administration forcing plants to stay online and trying to fund the construction of new coal plants, it’s worth understanding what drove so many plant operators to cease burning coal and switch to natural gas.

Drivers of coal-to-gas conversion

The spate of coal-to-gas conversions that began around 2008 was the product of two factors.

The first was regulatory. Burning coal emits a great deal of harmful pollutants (such as mercury), and over time regulation of these emissions has become stricter. In 2000 the Environmental Protection Agency (EPA) decided to develop regulations for the emission of mercury, and while this was temporarily delayed by the Bush Administration, by 2008 it was clear that stricter coal plant emissions would be a reality. In 2011, the EPA proposed a new set of coal plant emissions restrictions, the Mercury and Air Toxics Standards (MATS), which dramatically reduced the amount of mercury, toxic metals, and acid gases that coal and oil plants were allowed to emit. MATS, however, didn’t apply to natural gas plants, as gas burns much more cleanly and produces dramatically less harmful particulate emissions.

Alongside this new, more stringent regulation, the US shale gas boom made natural gas an increasingly attractive fuel for generating power. Between the late 1980s and 2011, natural gas went from 10% to nearly 30% of US electricity generation. And while the price of gas had risen through the early 2000s, it began to fall steeply in 2008. What’s more, it was projected to stay cheap for the foreseeable future.

Faced with increasingly strict environmental regulation and the rise of widely available and affordable natural gas, coal plant owners were faced with several options. One was to simply shut down their plants. Another was to install the required equipment to reduce emissions enough to comply with MATS regulations. This equipment was expensive to install and acted as a drag on plant efficiency, since it took energy to operate, but it was nevertheless often worth it. Today there are 219 operating coal plants in the US, all of which are in compliance with the original MATS regulations.

But some plant operators, instead of installing the required emissions equipment to comply with MATS, opted to convert the plants to burn natural gas. Converted plants were generally older, smaller-capacity plants that were relatively inefficient, used to provide extra capacity when needed rather than supplying baseload power.

Converting coal plants to burn natural gas wasn’t a new idea — the idea first began to be discussed in the 1980s, and during the 1990s and 2000s a few plants were converted — but the shale boom combined with the new, more stringent MATS regulations created a much stronger incentive to do so.

Converting a coal plant to burn natural gas

There are a few different ways of converting a coal power plant to a natural gas plant, depending on how much of the original equipment you replace. At a high level, a coal plant consists of a boiler that burns coal and uses that energy to turn water into steam, a turbine which converts the heat energy of the steam into rotational energy, and a generator connected to the turbine which converts the mechanical rotational energy into electric current. Different conversion strategies replace different portions of this equipment.

The simplest, cheapest option is to convert the coal boiler into one that’s capable of burning natural gas. The boiler can be converted to run on only natural gas (losing its ability to burn coal) or be converted such that it can burn gas, coal, or some combination of the two.

At minimum this sort of conversion requires adding a system for delivering natural gas into the boiler and replacing the coal burners with natural gas burners. But because natural gas burns differently than coal, this may also require other upgrades, such as new flame scanners (which monitor how combustion is taking place in the boiler) and structural upgrades (because the temperatures in certain parts of the boiler might be higher when burning natural gas). It also often requires adding a natural gas pipeline to bring gas to the plant, which might involve laying 20 miles or more of underground pipeline. Because natural gas doesn’t require the sort of complex material handling that coal does — it can simply be piped directly into the boiler — the upgraded plant typically requires many fewer employees than the original coal plant did. The Joliet coal plant near Chicago, which was converted to a natural gas peaker plant in 2016 before being shuttered entirely in 2023, is an example of this sort of coal-to-gas conversion.

The benefit of this sort of conversion is that it’s comparatively simple and inexpensive, and most coal-to-gas plant conversions in the US have been of this type. The drawback is that because the boiler is burning a fuel it wasn’t originally designed for, the converted plant operates less efficiently and with less capacity than it did while burning coal, and much less efficiently than a brand-new combined-cycle plant would.

Another option is to replace the entire coal boiler with a natural gas boiler, while keeping the rest of the generation equipment. This is less common, but it does happen. An example of this sort of conversion is Iowa State University, which operated a small 46-megawatt combined heat and power coal plant. In 2016 the university replaced three of the plant’s five coal-fired boilers with natural gas boilers. (The last two boilers were eventually also modified to burn natural gas).

More common than just swapping out a coal boiler for a gas boiler is replacing the coal boiler with a gas turbine and a gas boiler (which in this configuration is called a Heat Recovery Steam Generator, or HRSG), to create a more efficient combined-cycle plant. An example of sort of conversion, which is sometimes called “repowering,” is the Big Bend coal power plant near Tampa, Florida, which was converted in 2023.

And finally, a “conversion” can also simply rip out the entire existing plant — boiler, turbine, generator — and replace all of it with a modern combined-cycle plant. This is essentially a brand-new plant that uses some of the services (the grid interconnection, water availability) as the old plant. The Tennessee Valley Authority’s Allen plant near Memphis, Tennessee is an example of this sort of conversion. This type of conversion is the most expensive, but it gives operators the greatest increase in efficiency. Roughly a third of coal plant conversions in the US have been this sort of total replacement.

The future of coal-to-gas conversions

It doesn’t seem likely that we’ll see many more of these coal-to-gas conversions. For one, there are just a lot fewer coal plants in the US than there used to be: we’re down to just over 200 from a peak of nearly 600. The most obvious candidates for conversion — smaller, older plants that might be useful for peaking — have probably already been converted. And as grid-scale batteries change the economic logic of peaking, even new gas plants are looking less attractive than they used to; I can only imagine that a less-efficient converted coal plant is even less compelling.

It’s possible that a new round of more stringent air pollution regulations might push some of these existing plants into burning natural gas rather than coal, though it’s unlikely we’ll see such a thing during the Trump Administration. (In 2024 the Biden Administration strengthened the original MATS rules, but these additions have since been repealed by the Trump Administration.) Similarly, the enormous demand for power caused by the AI boom might have some effect. We’re already seeing coal plants slated for shutdown staying online instead due to high power demand and the huge backlog for things like natural gas turbines. It’s possible such logic might incentivize converting some remaining coal plants into burning natural gas instead. But overall, I suspect that the heyday of coal-to-gas conversions is behind us.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at homes being built on top of libraries, Patriot missile manufacturing, an effort to construct new US coal plants, a tunnel between the US and Russia, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housing

Apparently the hot new trend is to combine housing developments with libraries, either by building housing on the site of an existing library, or building a new library as part of a larger development project. “The projects would be part of a nationwide trend, joining similar efforts in the suburbs of Washington and the cities of Boston, Cleveland and Milwaukee. An October 2025 report from the Urban Institute, a nonprofit research group, found that more than 1,800 apartments had been built across the United States as part of developments combining libraries and housing. A number of such projects have moved forward in New York City, including at the Sunset Park Library in Brooklyn, the Inwood Library in Manhattan and the Grand Concourse Library in the Bronx.” [NYT]

Claims about housing construction in Taiwan. [X]

Manufacturing

It apparently takes more than two years(!) to build a Patriot missile. The article is thin on details, but this mostly seems to be due to the time suppliers need to make individual parts and subcomponents. [WSJ]

Claims that on the order of 90% all the semiconductor lithography equipment that’s ever been made is still in use. [X]

Chinese car manufacturer BYD wants to unseat Toyota as the world’s biggest carmaker in five years. [The Guardian] And in spite of the sky-high tariffs on Chinese vehicles, some Chinese car manufacturers have apparently established a small presence in the US in anticipation of future US sales. [Jalopnik]

The US adds a bunch of major Chinese manufacturers, including car manufacturer BYD, robot manufacturer Unitree, and battery manufacturer CALB to its list of “Chinese military suppliers,” which can make it more complicated for US companies to do business with them. [CarNewsChina]

Ford is trying to get more people into the skilled trades. [WSJ]

Google orders 3 million TPUs from Intel. [Tom’s Hardware] And SemiAnalysis thinks that things for Intel are looking up, and the company should raise capital. [SemiAnalysis]

Energy

Solar energy supplies more electricity than coal in the US for the first time. “Solar supplied 12.8% of US electricity last month while coal accounted for 12.2%, according to a report Wednesday from the clean energy think tank Ember, which analyzed monthly and hourly data from the US Energy Information Administration.” [Bloomberg] At its peak in the 1980s, coal supplied roughly 55% of US electricity.

But even though coal is in long-term decline, it’s not going down without a fight. The data center boom is keeping coal plants that were slated for closure operating longer. [The Register] And the Trump administration wants to use the Defense Production Act to build two new coal plants in Alaska and West Virginia. [Bloomberg] There hasn’t been a utility-scale coal plant built in the US since 2013.

Read more

Real human technology is, of course, not generated by randomly combining components together and seeing if they do anything useful; the randomness in these simulations is just a way to see how easy or hard it is to create new technologies under different conditions. But biological technology — the huge panoply of lifeforms that exist on earth, from microscopic single-celled organisms to whales the size of a 737 — is also generated by randomness. Evolution builds biological technology bit by bit by harvesting the fruits of genetic variation, often caused by random mutation, preferentially selecting the most fit organisms to propagate their genes into the future. Over billions of years, this process can generate astoundingly complex biological systems.

What’s interesting is that biological evolution uses a very similar trick to Arthur’s circuit simulation. By leveraging modularity at the genetic level, populations of organisms can increase the rate that useful genetic variants spread through the population, effectively increasing their rate of information acquisition. Sexual reproduction, along with other ways of sharing genetic material like horizontal gene transfer, is essentially a mechanism for doing this. We can show this with some simple simulations.

Evolution and reproductive strategies

The simplest way for an organism to reproduce is asexual reproduction, where a parent produces a child that’s a genetic copy of itself. Simple single-celled organisms, for instance, reproduce by cellular fission, dividing into two or more “children” that each have the same genes as the original parent.

But children won’t necessarily be identical copies of their parents. Due to genetic mutation, some genes might get randomly altered during the fission process, producing children with slightly different genes. In some cases, these mutations might be useful, giving additional functionality such as antibiotic resistance and thus better odds of surviving and reproducing. Because of their contribution to the organism’s fitness, over time the useful mutations will become more and more common in the population.

We can demonstrate this with a simple simulation. In our simulation, we start with a population of 100 creatures, each of which has a genome of 200 individual genes. A gene can either be a 1 (the “good” version of the gene) or a 0 (the “bad” version of the gene). The initial population is random, with each creature having roughly a 50-50 mix of good and bad genes. Each iteration of the simulation, each creature produces two children. A child copies the genes of its parent, but due to mutation each gene has a 0.2% chance of being flipped, going from a 1 to a 0 or vice versa. The 100 most fit children (where fitness is just the sum of each gene value, since 1 is the “good” version of the gene in our simplified model) are selected to continue the next generation, and the cycle repeats. This is a simplification compared to how evolution actually functions — for one, it treats genes as contributing to fitness independently, ignoring the fact that the fitness value of one gene often depend on other genes — but it’s enough to show some of the dynamics at work.

When we run this simulation, the proportion of “good” genes in the population steadily rises over time as more-fit offspring outcompete less-fit offspring. Depending on the mutation rate, the population may eventually reach maximum possible fitness of 200, or plateau at some level below it.

The problem with this strategy — producing children that are noisy copies of a single parent, and relying purely on random mutation as a source of genetic variation — is that once you’re at above-average fitness, mutations are likely to be bad on average. If a genome has more 1s than 0s, a random change will be more likely to change a 1 to a 0 than a 0 to a 1. Thus for parents of above-average fitness, their children will on average have lower fitness.

Because mutation is random, there will nonetheless be variation, and some children will end up with higher fitness than their parents. And because selection eliminates the least fit each iteration, the pool of selected children will have higher average fitness than their parents, allowing average fitness to increase over time. But mutation reducing average fitness drags down this process.

You can see this in the graph below, which shows a simulation with slightly different parameters (a genome length of 1000 and a mutation rate of 2%) to more easily see the trends. The top graph shows the distribution of population fitness at generation 50, and the second graph shows the distribution of the population’s children prior to selection. You can see that, thanks to mutation, average fitness has dropped, though due to randomness some proportion of the children have lucked into getting higher fitness. The last graph shows the children after the top half of the distribution has been selected. Average fitness rises, and is now above the initial population, though just barely.

Now let’s look at a simulation of a different reproductive strategy: sexual reproduction, where children get their genes from two parents rather than just one. In this simulation, we still have a population of 100 creatures with genomes of 200 genes, each of which can either be a 0 or a 1. But now children have two parents, and in each iteration members of the population are paired up randomly and each pair produces four children. Children get their genes from both parents with each gene having a 50% chance of coming from a particular parent. The top 100 most fit children are then selected for the next generation, and the iteration continues. In this simulation, there is no mutation, so genetic variation entirely comes from reshuffling the genes of the parents.

Like the previous simulation, the population gradually reaches maximum fitness. But sexual reproduction gets there much faster. With asexual reproduction, after 200 generations the population was around an average fitness of 187. With sexual reproduction, the population average reached the fitness maximum of 200 in just 33 generations.

The key is that sexual reproduction introduces genetic variation without reducing average fitness. Since children are a random combination of their parents’ genes, on average they’ll have the same fitness as their parents (with some randomly having higher fitness, and others randomly having lower fitness). When the most-fit children are selected for the next generation, this is taking the top half of a distribution with a much higher average than the distribution of children in the asexual simulation. Average fitness thus rises much faster.

If you work out the math (or, as I did, simply read the math that someone else worked out), in an asexual population the rate of fitness increase is 1/(8*f), where f is the differential normalized fitness. (The normalized fitness of a population is the average fraction of good genes in that population; so a population where, on average, a member has 150 good genes in a genome of 200 would have a normalized fitness of 0.75. Differential normalized fitness is the normalized fitness of the population minus 0.5, the normalized fitness of a randomly generated population.) Early on, fitness of a population can increase quickly, but the rate soon drops below an increase of 1 unit of fitness per generation (one gene flipped from a 0 to a 1 on average). As a population gets closer to maximum possible fitness, the rate of fitness increase approaches 0.25 (flipping one gene, on average, from a 0 to a 1 every four generations).

With sexual reproduction, on the other hand, the rate of fitness increase turns out to be much higher: it’s proportional to the square root of the length of the genome.

The informational power of genetic recombination

One way to think about why sexual reproduction is so powerful is to look at lineages of descent. Say that one of the members of our asexually reproducing population stumbles across a new, useful mutation. Because genes are passed from one parent to one child, the only way this gene can spread throughout the population (in the absence of some other member of the population also stumbling across it) is if the children of whoever has it outcompete the children of everyone else. In this scenario, the population eventually ends up consisting entirely of descendants of one particular member of the population — as a necessary condition of this spread, every other genetic lineage (along with whatever useful mutations they might have stumbled across) gets wiped out.

We can see this in our simulation results. The chart below assigns each member of the initial population, and their children, a unique color. The simulation starts out with 100 different colors (one for each member of the population), but this quickly gets winnowed down to a much smaller number. After a few generations, the population is one uniform color, all descendents of one particular member of the initial population. (This chart is from one particular simulation run, but repeated runs will show the same behavior.)

If we redo the color coding whenever the population reaches the point where everyone is descended from a single ancestor, we see that this happens repeatedly. In the chart below, the population at generation 48 are all descendants of one particular member of the population who lived in generation 25. At generation 80, they’re all descendants of one particular member alive from generation 48.

In evolutionary biology, this phenomenon is known as “clonal interference”: if two different beneficial mutations arise in different members of the population of the same generation they can’t be shared and so they end up competing against each other, and one beneficial mutation ultimately gets wiped out.

With a sexually reproducing population, on the other hand, useful mutations can be shared much more easily. In an asexual population, a member has one parent, one grandparent, one great-grandparent, and so on. But in a sexual population, a member has two parents, four grandparents, eight great-grandparents, etc. Beneficial variation from earlier generations can spread much more easily.

We can see this in the graph below, which shows the proportion of genes of the original members of a sexually reproducing population represented in the gene pool at any given time. We can see that the proportion stays very high: genes from almost 75% of the original population are found in the population after 34 generations. In the asexual population, this was 1% (and would be even lower in larger populations, since it’s just 1/total starting population).

We previously noted that Brian Arthur’s circuit simulation took advantage of modularity, finding useful subcomponents, locking in their designs, and then using those to build more complex technologies. Once the simulation finds a 3-way AND gate, it can use that to make a 4-way AND gate, which it can use to make a 5-way AND gate. We likewise noted that if you’re trying to build an 8-bit adder by randomly combining NAND gates together, it’s vastly easier if you can add one NAND gate at a time and verify you’re correct than if you have to guess all 68 gates at once.

You can think of this like someone solving a combination lock. A lock with a five-digit combination, with 100 possible values for each digit, has 100^5 = 10 billion possible combinations. Trying combinations one by one would take forever.

Technological modularity is like being a skilled lockpick that can check whether each individual digit attempted on a combination is correct (maybe by listening carefully you can hear a telltale “click” when the dial is in the right spot). Now instead of searching over 10 billion possible combinations, you’re doing five searches over 100 possible values each, or 500 possibilities total. The space of possible options that must be considered is vastly reduced.

Sexual reproduction is, as I understand it, doing something similar: by letting genes from two parents be combined to form children, it effectively lets the fitness of each gene be tested independently, turning the search from something like “find the best 200-gene genome” to something closer to 200 parallel “find the best gene at this location” searches. In our combination lock analogy, the modular circuit simulation is sort of like turning the dial until you hear a “click,” which indicates that the given number is correct. Sexual reproduction is more like trying a bunch of different random combinations, getting back a score for “how close this combination is to being solved,” and using that to infer which “dials” are correct. The search space is correspondingly greatly narrowed, and the search proceeds much faster.

With Arthur’s technological search, we couched this reduction in terms of information theory, calculating the bits of information gleaned per iteration. (As a reminder, a “bit” is just “something that cuts the number of possibilities to consider in half.”) In our 8-bit adder, 68 NAND gate search, finding the working arrangement required narrowing down 2^853 possibilities, or 853 bits. Trying to get all 68 gates right at once got us less than 0.000001 bits per attempt, requiring a long and painful search. But with modularity — going gate-by-gate and knowing whether each gate is in the correct location — we accumulate information much faster, around 0.003 bits per attempt (3,000x faster than nonmodular search).

We can similarly look at biological evolution and the spread of useful genetic variants in terms of information acquisition. For our simulations, the information we have at a given time is a function of how “certain” the population is about the value of each gene. With our starting, randomized population for each gene approximately 50% of the population has a 1 and 50% has a 0, we’re maximally uncertain, and we have 0 bits of information for each gene. When the population has reached maximum fitness (every member having a 1 for every gene) we’re maximally certain, and we have 1 bit of information for each gene. Total information is thus roughly equal to 2 * (F - G/2), where F is fitness and G is genome length.

We can see that information is acquired much more quickly in our sexual reproduction simulation than in our asexual simulation.

There’s an important caveat to all the above analysis: it assumes that genes independently contribute to fitness: that is, that the usefulness of gene 27 isn’t a function of gene 145. When genes are coupled like this, and the usefulness of one genetic variant is a function of what other genetic variants you possess (as they often are in practice), the evolutionary search process gets much more complex to analyze. (Stuart Kauffman has done a lot of work here with his concept of NK landscapes, which you can read about here and here.) But the simple, additive fitness case is still useful for understanding these evolutionary mechanics.

It’s also important to know that in real life, asexually reproducing organisms like bacteria have ways of sharing genes between them that get them some of the benefits of sexual reproduction. Bacteria widely engage in what’s known as horizontal gene transfer, which is exactly what it sounds like: genes being transferred between existing members of a population. This is apparently how genes for antibiotic resistance mostly spread, and some analyses suggest that 20-80% of bacterial genomes are the result of this sort of gene transfer.

Conclusion

The technological search process benefits greatly from modularity: being able to break a technology down into subcomponents with specific functionality, and determining whether those subcomponents are functioning properly. In information theoretic terms, this greatly narrows the possibilities that must be considered in a search process. It’s interesting to see that biological evolution, which creates and operates in an entirely separate domain, uses a similar sort of trick: using genetic recombination (in the form of sexual reproduction and horizontal gene transfer) to make the search process more modular and gain information more rapidly.

(For more about these ideas about evolution and information acquisition, including a much more rigorous mathematical treatment, see chapter 19 in David Mackay’s book “Information Theory, Inference, and Learning Algorithms.”)

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at chatbots replacing realtors, Chinese synthetic diamonds, Australian batteries, Meta’s data center tents, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Iran war

Iran breaks off negotiations with the US and vows to “completely block” the Strait of Hormuz. [CNBC]

Analysis of satellite data by the BBC suggests the damage Iran has inflicted on US military facilities is more extensive than has been previously reported. [BBC]

Housing

A NYT reporter successfully uses an AI chatbot instead of a realtor to sell their house. “A flurry of bookings to view our house over the coming weekend arrived in my inbox within hours. I struggled to manage the appointments until, again, I just let the chatbot do everything for me. I told agents that they had to email or text — no phone calls. Whatever they wrote, I copied and pasted into the chatbot; whatever it replied, I copied and pasted right back to the agents. I was worried that pushy ones would prey on my inexperience, so I had the chatbot come up with a list of potential conflicts and write confident responses I could have ready.” [NYT]

Opposition to property taxes is having a political moment, which as we’ve mentioned previously is a pretty bad idea. Now Florida really seems like it might be on the verge of effectively eliminating homeowner property taxes. “Gov. Ron DeSantis’ property tax plan for the November ballot would raise the homestead exemption to $250,000 and require the Legislature to enact a plan to eliminate property taxes entirely for the vast majority of Floridians who own the homes they live in, he announced Wednesday DeSantis said he was calling the Legislature back to Tallahassee on Monday to add an amendment to the ballot that would eventually eliminate property taxes for 92% of those Floridians by raising the homestead exemption to $500,000.” [Governing]

The urban benefits of allowing tall buildings. “Land-use regulations, including height limits, affect housing affordability and urban productivity. This column analyses over 11,000 urban agglomerations and 300,000 tall buildings to explore the effect of height restrictions on welfare. Vertical growth enhances land efficiency, reduces commuting, and boosts worker welfare. While higher density can increase housing demand and rents, the associated gains more than offset the costs.” [VoxEU]

It’s apparently easier to get planning permission to build a skyscraper in London (a city which has notoriously made it almost impossible to build new housing) if you include a publicly accessible roof deck, and thus quite a few London skyscrapers have them. [Diamond Geezer]

A census map of where air conditioning is uncommon in the US. [X]

Manufacturing

As AI gets more capable, the risk that someone uses it to design an engineered virus or pathogen rises. To help mitigate this risk, IFP co-led an open letter calling for mandatory record-keeping of who has purchased synthetic nucleic acids and the equipment for making them. “As life sciences researchers, builders of AI and biotechnology, and experts with a wide range of views on how to approach AI policy, we call on legislators to make screening of orders for synthetic nucleic acids — and the equipment needed to make them — mandatory.” [ScreenDNA.org]

One surprising beneficiary of the AI boom: Chinese manufacturers of synthetic diamonds, which are used as part of cooling systems for semiconductors. “Traditionally associated with jewelry, these synthetic gems are now being adopted as chip‑cooling materials, enabling denser and more powerful AI semiconductors. Momentum has accelerated after several Chinese producers reported that clients validated their diamonds as effective heat spreaders and began commercial shipments.” [Bloomberg]

California passes a bill requiring 3D printers to include controls to block the production of “ghost guns.” [The Register]

Elon Musk’s “Terafab” semiconductor fab gets a big tax break from a Texas county. [KBTX]

Read more

Many folks, including me, have observed that it seems to take much longer to build infrastructure in the US than it used to. People point to things like the rapid construction of the Empire State Building (one year) or the Golden Gate Bridge (just over four years) and note that for a modern infrastructure project it can take that long or longer to even get the permits or do the environmental studies.

But when doing these sorts of comparisons, it’s important to compare like with like: specifically, we shouldn’t measure the time spent planning a project (which would include doing the environmental studies and securing the permits) against the time spent actually building it. (The Golden Gate Bridge, for instance, was constructed in four years from 1933 to 1937, but planning for the project began around 1921.)

I wanted to get a sense of how planning times for major infrastructure projects in the US have evolved over time. To do this, I looked at planning and construction times for 67 major bridges built in the US since roughly the beginning of the 20th century. For each bridge, I noted the year that planning began, the year that construction began, and the year that it opened for service.1

As usual with an exercise like this, the results you get will be a function of the definitions you choose. “Started construction” and “opened for service” are relatively unambiguous — but what specifically do we mean by “planning begins”? People will often float the idea of a bridge for years before anything resembling formal plans is in place; the idea for the Brooklyn Bridge, completed in 1883, apparently dates back to 1804, but it doesn’t seem reasonable to consider that early discussion the beginning of planning. On the other hand, if you choose something like “the government formally announces the project” as your criterion, that might exclude years of serious efforts to get a project constructed.

I ended up choosing the “planning begins” date as the point when some organization connected with transit planning in some official capacity first announced or proposed the project. So for the Golden Gate Bridge, this would be 1921, when Chicago engineer Joseph Strauss and San Francisco city engineer Michael O’Shaughnessy prepared a joint proposal for the bridge. However, pinning these dates down was often difficult. Most bridges aren’t the Golden Gate Bridge, with lots of publicly available sources chronicling their design and construction. Often sources give a brief “planning begins” date without explaining what that meant specifically or what activities preceded it. And for some bridges, such as the New Tappan Zee Bridge, the decision to proceed with a bridge was preceded by years of studies on possible ways to expand transit capacity that considered many non-bridge options. For these, I tried to use the initiation of broader studies as the start of planning, but I suspect that these sorts of preliminary studies weren’t always documented. Overall I did my best to determine the dates, but I expect there to be errors and inconsistencies, and the results below should be taken with a grain of salt.

Those caveats out of the way, let’s look at the data. The graph below shows planning times for each bridge on the list. The horizontal axis is the year when planning began for a particular bridge, and the vertical axis is total years of planning.

And the graph below shows construction times for each bridge.

Because there’s a lot of variation in both planning times and construction times, it’s useful to look at trends by age bracket. The chart below shows average planning times and construction times for 20-year age buckets, with pre-1900 and post-2000 bridges each given their own bucket.

There are a few trends visible here. One is that bridge construction times fell from 1900 to the 1960s, but since then have risen. The average time to construct a bridge between 1980 and 1999 (6.5 years) was more than twice as long as the average construction time for bridges built between 1940 and 1959 (3.1 years). Construction times for bridges built between 2000 and 2025 are down from this peak (5.4 years), but they’re still well above the times of the 1920s through the 1950s.

Planning timelines, on the other hand, show a somewhat different trend. At a high level, we do see something that looks like the “times fall, then rise again” pattern we saw with construction times: planning times fell in half from over 12 years between 1900 and 1919 to 6 years between 1960 and 1979, before rising again in the 1980s. But the 1980-1999 period planning times look fairly similar to every other period other than 1960-1979. It’s not clear that modern planning timelines, which are on the order of 10 years on average, are all that different from most historic ones.

More recently, post-2000 planning times look even shorter (6.2 years on average), but this data is almost certainly biased downward because the bridges with long planning times haven’t been completed yet. A bridge that started planning in 2007 but has a 20-year planning timeline won’t start construction until 2027, and thus won’t show up in the data. A similar thing is probably taking place for pre-1900 bridges, which have planning times which are biased upward (since I was mostly picking bridges that were built after 1900). So we should be careful about drawing conclusions about planning times for these age brackets.

One notable thing about recent bridge construction is that a very large fraction of modern bridges are replacements for existing bridges. Of the 17 bridges on the list that were completed after the year 2000, 14 of them were replacement bridges. Some of these were for bridges that collapsed, like the I-35W bridge replacement in Minneapolis; others were for bridges that were damaged and at risk of collapse, like the Penobscot Narrows Bridge in Maine; and some were for bridges that were near the end of their design lives, like Florida’s Pensacola Bay Bridge. All else being equal, I’d expect a replacement for an existing bridge to get built more quickly than a completely new bridge, because people will react more viscerally and negatively to a major traffic route being removed than they will for construction of a nonexistent bridge getting delayed. This visible sentiment probably creates a sense of urgency to replace an existing bridge that doesn’t exist for a completely new bridge that people haven’t developed expectations for. One illustration of this dynamic is that collapsed highway overpasses often get rebuilt exceptionally quickly. When a highway overpass in Atlanta collapsed following a fire, it was rebuilt in just six weeks. We saw something similar with collapsed highway overpasses in Los Angeles and Pennsylvania. (The long time its taking to rebuild the collapsed Francis Scott Key Bridge in Baltimore thus may be something of an outlier.)

One other notable observation: there appears to be little relationship between planning times and construction times. Bridges that take a long time to plan might get built exceptionally quickly, and vice versa.2

Which bridges on the list took the longest to plan? Here are the bridges with the five longest planning timelines.

• Bear Mountain Bridge, New York, 54 years — The Bear Mountain Bridge, a suspension bridge in New York, was first planned in 1868, after a bill was passed by the New York legislature. There were a series of attempts to construct the bridge over the next 30 years, but none of them was successful in building more than foundations, and construction on the current bridge didn’t begin until 1922.

• Fremont Bridge, Oregon, 41 years — Planning for the Fremont Bridge in Portland dates back to 1927, when the City Council issued a “Major Traffic Street Report” which proposed a bridge across the Willamette River in the Fremont Bridge’s current location. The bridge was also suggested by the Oregon State Transportation Committee in 1933 and by Robert Moses in his 1943 planning study for the City of Portland. The site for the bridge was finally settled on in 1955, and construction began in 1969.

• Queensboro Bridge, New York, 34 years — The earliest proposals for a bridge between Manhattan and Queens date back to 1804, but the first serious efforts began in 1867, when wealthy Queens residents formed the New York and Long Island Bridge Company to build the bridge. The company hosted a design competition for the bridge in 1876, but work was slow to begin, and in 1890 a state justice ruled the charter invalid. In 1900, the city took over the company’s franchise to build a bridge, and construction began in 1901.

• Claiborne Pell Newport Bridge, Rhode Island, 32 years — Planning for the Claiborne Pell Newport Bridge in Rhode Island began in 1934 but was delayed until after World War II for unclear reasons. In 1948 the state created the Newport-Jamestown Civic Commission to explore ways to finance and construct the bridge, but construction didn’t begin until 1966.

• Verrazzano Narrows Bridge, New York, 31 years — In 1928, the chambers of commerce for several New York boroughs announced that the Interboro Bridge Company had proposed a “Liberty Bridge” that would connect Brooklyn and Staten Island. However, a vote on the proposed bridge in the New York legislature was blocked by congressman and future New York mayor Fiorello La Guardia. This was followed by plans for a tunnel (which were then shelved), then another bridge, then another tunnel, until finally the plans for a bridge were approved in the late 1950s. Construction began in 1959.

Conclusion

Planning timelines for bridge construction in the US have not gotten obviously worse over the past several decades, though I suspect that the fact that most recently constructed bridges are replacing existing bridges is doing much of the heavy lifting here. This supports a vague notion I have that organizational success is often downstream of a culture of urgency — considering a problem important to actually solve and being willing to work hard on solving it. (You get the same sense of urgency when reading about how countries like Japan and Korea were able to build their commercial shipbuilding industries up from virtually nothing.) Perhaps the trick to making US infrastructure construction proceed more quickly is figuring out how to instill this culture of urgency in the agencies responsible for building it rather than relying on imminent infrastructure failure to create it.

House building using the “Oraaflex” modular construction system, via Wikipedia.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at a California chemical leak, weapons-grade plutonium for nuclear reactor startups, a startup that will clean your house to get robot training data, Blue Origin’s rocket explosion, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housekeeping items:

• I spoke at the Reagan National Economic Forum about energy and AI infrastructure. You can watch my panel here.

War in Iran

Iran ends 88-day long internet blackout. [The Hill] And Iran claims that the fees it’s charging to allow ships to pass aren’t tolls but “safety and environmental protection fees.” [MSN]

Housing

West coast cities such as Seattle and Portland are considering a “vacancy tax” to try and deal with the problem of empty commercial office space. Such taxes already exist in Washington DC and San Francisco. [The Urbanist]

Relatedly, developer Asher Luzzatto is trying to revitalize Denver’s downtown, which currently sits at a 40% office vacancy rate. “Luzzatto’s approach is to turn that old formula inside out, transforming desolate urban cores into welcoming places to live. He plans to convert half of the Energy Center plus two other downtown Denver office buildings into about 1,100 apartments. He’s also planning a bookstore, art gallery, children’s museum and daycare center. He acquired all four buildings for pennies on the dollar. In a matter of months, the Los Angeles native has taken control of more than 7% of Denver’s traditional downtown office space.” [WSJ]

What was it like to live in an apartment in ancient Rome? [Common Edge]

Manufacturing

In California, damage to a tank of methyl methacrylate (a substance used to make plastic) put 50,000 people under an evacuation order due to the possible risk of an explosion. Fortunately the problem was resolved without the tank exploding. [BBC] At “In the Pipeline,” chemist Derek Lowe talks about methyl methacrylate and how it could potentially cause an explosion. “There’s a key energetic aspect to these polymerizations: they’re thermodynamically favorable, and these bond formations give off a bit of heat as they occur. This heats up the solution as a whole, and that in turn speeds up the reactions all by itself! Which means even more heat as even more molecules polymerize. . .and now you see why storage of large quantities of these monomer compounds is not for the unwary. You have to keep them away from anything that can start a free-radical chain reaction, and that means light and heat for starters, but also not letting them sit in contact with many metals and alloys, etc.” [Science]

In other “damaged chemical tank” news, one person was killed at a chemical tank explosion at a paper mill in Washington state. [USA Today]

Nvidia plans to spend on the order of $150 billion in Taiwan each year. [Reuters]

Chinese tech firm Huawei claims a major chip design breakthrough. “Huawei said Monday at a tech conference in Shanghai that by 2031, its high-end chips would have transistor density equivalent to 1.4-nanometer processes, which is considered the industry’s cutting edge.” [MSNBC] And Chinese battery manufacturer Calb is building a battery factory in Portugal that will produce on the order of 4% of the country’s GDP. [X]

A report on the state of US missile manufacturing. “America’s missile production hinges on a small number of ammonium perchlorate facilities, meaning a single plant accident can bring output to a standstill; a concentration risk that has no real equivalent elsewhere in the defense industrial base. AP production relies on narrow workforce pipelines for specialized energetics handling, layered environmental and explosives permitting, and purpose-built manufacturing equipment. Each of these inputs is hard to duplicate quickly, even with dedicated funding, which is why decades of rhetoric about supply chain expansion have produced so few second sources in practice.” [Contrary]

Inside Ukraine’s efforts to manufacture thousands of ground-based drones. [TWZ]

Read more

Now that we’ve mapped the contours of the problem, we can begin to explore its deeper nature to understand why, specifically, construction productivity is so resistant to being improved, and why construction costs stubbornly refuse to fall.

We’ll start by looking at one of the most important mechanisms by which production processes can get cheaper: economies of scale. Many processes have lower unit costs as production volumes rise, thanks to a variety of scaling effects: fixed costs can be spread more thinly, equipment gets cheaper on a per-unit basis due to area-volume relationships, improved production methods are developed as a result of learning-by-doing, and so on. However, in construction these effects are modest at best, even in sectors like homebuilding with very large production volumes.

In homebuilding, we’ll see that the limits to these economies of scale are in large part dictated by the nature of the production process. Economies of scale work by eliminating the difference between the costs of the raw inputs to a process and the final costs of production. In a highly efficient, high-volume production process, the costs of the output will gradually approach the costs of the material inputs. But in conventional homebuilding in the US, this difference is already small, giving scale-based strategies relatively little margin to close.

Economies of scale in homebuilding

We’ll examine economies of scale in construction through the lens of housing construction in the US. For many sectors of construction, difficulty in achieving economies of scale could be attributed to the fact that only a small number of buildings of a particular type get built in the US each year. There were, for instance, only 10 skyscrapers taller than 200 meters built in the US in 2025. Semiconductor fabs, urban subways, and airports are similarly built in very small numbers. It’s hard to achieve economies of scale when there’s no scale to be had. Houses, on the other hand, are built in very large numbers. There were over 1.3 million housing starts in the US last year, including 942,000 single-family homes. This isn’t large compared to other types of manufactured goods — the US consumed over 7 billion cans of vegetables in 2025, for instance — but it’s certainly large enough for economies of scale to appear.

However, evidence suggests that homebuilding in the US exhibits relatively modest economies of scale. For one, the level of concentration in the homebuilding industry is relatively small: housing construction in the US is done by a large number of comparatively small firms. A 2022 study by Harvard’s Joint Center for Housing Studies (JCHS) found that the US had over 65,000 firms engaged in homebuilding, and that even the 100 largest firms combined were responsible for less than 50% of the homebuilding market. In homebuilding, the four largest firms held around 18% of the market, compared to 90% in aircraft manufacturing, 86% in wireless telephone service, and 58% in automobile manufacturing. Concentration in the homebuilding industry has been rising over time (primarily due to the growth of the two largest builders, Lennar and D.R. Horton), but it’s still much lower than in many other industries. This level of concentration isn’t what you’d expect to see if economies of scale in homebuilding were substantial.

Another, earlier JCHS study that looked at US homebuilding in the late 1990s and early 2000s found that construction costs were actually higher for the largest US homebuilders than for smaller US homebuilders. It also found that large and small homebuilders had similar gross margins on homebuilding (the difference between the costs to produce a home and what they sold it for). If there were major economies of scale in homebuilding, all else being equal we’d instead expect to see larger firms have lower construction costs and greater gross margins.

More recent data suggest these trends have continued. If we look at the gross margins of public homebuilders, we see virtually no relationship between gross margins and number of homes produced. Lennar, which built over 80,000 homes in 2025, has similar gross margins and average home selling prices to United Homes Group, which sold just 1,192 homes in 2025. LGI Homes and D.R. Horton had nearly identical selling prices and gross margins, but the former sold 4,685 homes in 2025 and the latter sold over 84,000. Looking at different years doesn’t alter the picture much.

This isn’t a smoking gun, as there are other reasons why larger firms might not have higher gross margins than smaller firms. Large firms might be deliberately selling at a lower price to try to capture or maintain market share, as Lennar states it does. But it’s consistent with there being few economies of scale in the homebuilding industry.

In his history of the US homebuilding industry, homebuilder Ned Eichler describes the economies of scale available to homebuilders through the end of the 20th century. While he notes that homebuilders building several hundred homes per year could achieve substantial cost advantages compared to much smaller builders, there appeared to be little advantage in increasing scale beyond that. Writing in the 1980s, Eichler noted that “[n]ot even the largest firm has any of the advantages of dominant companies in other fields… What little technology exists is available to all.”

In the late 1940s and 1950s, as large-scale merchant builders (100 units per year or more) became the dominant suppliers of homes, most of the postwar improvements in production organization and techniques were made. As unit costs began to rise rapidly in the 1960s, because of higher quality, high land and site development costs, and inflation, there was heightened interest in the possibility of another round of increased efficiency from greater scale and technological breakthroughs. Many firms did grow, mostly by geographic expansion. Several broke out of the 100 to 1000 unit range to annual volumes over 5000. By 1980 US Home brought its unit production to 15,000. However, there were no significant changes in technology or even methodology. Nor is there clear evidence that US Home, Ryan, Pulte, or other multicity builders have yet achieved economies of scale.

Today, large homebuilders continue to use the same basic homebuilding methods that much smaller builders do, and builders like Lennar and D.R. Horton effectively subcontract all of their actual homebuilding to local contractors. (Lennar’s 2025 annual report, for instance, notes that “[w]e hire subcontractors for site improvements and virtually all of the work involved in the construction of homes.”) Large homebuilders do note that they’re able to achieve some scale economies by getting volume discounts on material purchases, but evidence suggests that these savings are relatively modest. Group purchasing organizations, such as CBUSA, pool material orders together in order to receive volume discounts from suppliers; CBUSA boasts that it gives independent builders “the purchasing power of a top 10 National Builder.”1 But the savings it offers works out to around $9,500 per house for the builders who use it, or around 6% of the material costs of an average new home.

You sometimes see claims that economies of scale in homebuilding could be achieved with assembly line-style construction, building hundreds or even thousands of homes at once on enormous tracts the way the US did in the 1940s and 50s. But evidence suggests that savings from this style of construction are somewhat modest, and may not materialize at all. Older versions of Craftsman’s National Construction Estimator provide percentage cost deductions for some tasks if they’re done on “tract work,” but these reductions are typically modest (10-20%) and are only available for a small handful of tasks.2 (One exception is carpentry, which does see large labor cost reductions for tract construction in some editions, but this on its own isn’t enough to shift the overall cost of a house substantially.) Likewise, Levitt and Sons famously built thousands of houses in enormous “Levittowns” using a sort of “reverse assembly line”: worker teams would go from house to house, performing a particular set of tasks and then moving on to the next one. But the cost of Levittown houses doesn’t appear to be all that different from those built by other, much smaller homebuilders at the time.

Economies of scale in manufactured homes

We can sharpen our understanding of economies of scale in homebuilding by looking at one particular subset of housing construction: manufactured homes. Manufactured homes, formerly known as mobile homes, are built using largely the same technology as conventional homes, but instead of being built on-site, the homes are produced in a factory, mounted to a steel chassis, and transported to their final location via truck. Roughly 100,000 manufactured homes are produced in the US each year.

Manufactured homes are a useful lens for understanding economies of scale in homebuilding, because they eliminate so many factors that might be expected to restrict them. Economies of scale require repetition — making approximately the same thing over and over again — and such repetition is particularly achievable with manufactured homes:

• Conventional homebuilding is subject to different building code requirements in different jurisdictions, depending on what version of the code has been adopted. But manufactured homes are built to one set of national requirements, the federal HUD code.

• Conventional homes are constructed on-site, and those sites might have substantial variation between them. But manufactured homes are produced repetitively within controlled factory locations.

• Conventional home construction takes place on constantly changing jobsites, with constantly changing construction crews, a constant churn that might disrupt opportunities for learning-by-doing. Manufactured home construction, which takes place within a single location, should be less susceptible to this sort of disruption.

There’s some evidence for greater economies of scale in manufactured home construction compared to conventional homebuilding. Specifically, the manufactured home industry shows substantially higher concentration than the conventional homebuilding industry. The three largest manufactured home producers are Clayton Homes (49,000 units per year), Cavco (about 20,000 units per year), and Champion (roughly 26,000 units per year).3 Together, they make up roughly 90% of the US manufactured home market, similar to the level of concentration we saw in other industries.

However, despite this level of concentration, actual production economies of scale in manufactured home building appear relatively modest. We can see this by looking at the operations of four public manufactured home companies. The first two, Champion and Cavco, produce around 26,000 and 20,000 homes per year, respectively. The third, Legacy Housing, makes around 1,700 manufactured homes a year. And the last, Nobility Homes, makes just a few hundred manufactured homes each year.

Each of these homebuilders is catering to broadly similar segments of the market, since manufactured homes tend to be purchased by lower-income buyers. Their annual reports all note that the typical manufactured home buyer has a low household income, though they disagree on the exact number – Cavco thinks it’s less than $40,000, Champion thinks it’s less than $60,000, and Legacy thinks it’s less than $75,000. As we saw with conventional homebuilders, there doesn’t seem to be much relationship between gross margins on factory homebuilding and production volume. Nobility homes (391 shipments in 2025) and Legacy homes (1703 shipments in 2025) actually have better gross margins on home sales (32% and 28% respectively) than the much larger Champion and Cavco (27% and 23%).

More importantly, all four of these manufacturers use essentially identical production methods. Nobility Homes has one factory in Florida, Legacy has two factories in Texas and one in Georgia, and Champion and Cavco each have several dozen factories spread across different states. These factories are similar: each is typically 100 to 200 thousand square feet, employs 100 to 300 people, and produces several hundred homes a year. In other words, there do not appear to be large economies of scale at the factory level. Even relatively modest production volumes (20,000 units annually) get spread among dozens of different factories. By comparison, a single modern car factory will produce several hundred thousand cars a year.

There are some exceptions to this: Clayton Homes recently announced renovations to an existing factory that will allow it to produce 3,000 homes annually. But overall Clayton fits the basic pattern, as production of its 49,000 homes is spread amongst 41 production facilities.

Within the manufactured home facilities, we see the adoption of various efficiency-enhancing strategies compared to how work would be done on a conventional jobsite. As with Ford’s assembly line, work is arranged so that tasks take place at worker height, eliminating extraneous movements like bending or stretching. With conventional home construction, workers first complete the outside shell of the house to create a waterproof environment to work in, but this means the remaining work must be done within the cramped interior of the partially completed house. Manufactured homes, built under a factory roof, aren’t subject to this restriction and can be built from the inside out. Permanent stations can be set up with equipment for various repetitive operations, such as cabinet assembly. (We see this same sort of strategy adopted in large-block ship construction, with as much work as possible done outside the ship on prefabricated blocks, minimizing the amount of work that needs to be done in the cramped conditions of the assembled ship.)

However, the work remains a highly labor-intensive operation. Workers are in large part doing the same tasks that they would be in conventional construction; those tasks have just been made somewhat easier to perform. Equipment use is likewise somewhat modest: we don’t see the huge arrays of robot arms or automated machinery that we see in car manufacturing or other mass production industries. Per their annual reports, the value of the equipment that Nobility Homes, Legacy Housing, and Cavco operate is just one-third to half the value of the actual buildings. At a car manufacturer like Ford, by contrast, the value of the equipment and tooling is over three times as much as the value of the buildings that contain them.

Part of the reason manufactured homebuilders use a large number of low-capacity factories is high transportation costs. Both Nobility Homes and Cavco note in their annual reports that the cost-effective shipping range for their products is around 350 miles (around a day’s drive). This is a result of dollar density — the value of the product divided by the volume or mass of it. Products with high dollar density, like electronics, are very valuable in proportion to the amount of space they take up and can be effectively shipped long distances; iPhones can be made in a factory in China and then shipped all over the world. Products with low dollar density, like toilet paper, have low prices in proportion to the amount of space they take up, and are harder to ship long distances. If it costs $2,000 to ship a container from China to the US, that $2,000 will be a small fraction of the value of the cargo if the container is filled with iPhones, but a large fraction of the value of the cargo if it’s full of lumber or kitty litter. The lower the dollar density of a product, the less cost-effective it is to ship it long distances. Manufactured homes have very low dollar density, and thus can’t be shipped very far cost effectively.

But transportation costs can only explain part of this distributed manufacturing strategy for manufactured homes. For one, if there were substantial economies of scale at work, it would be worth it for manufacturers to ship longer distances in spite of the higher costs; transportation costs are proportionately high for manufactured homes, but they’re not that high, on the order of a few dollars per mile. Transportation costs thus suggest a sort of ceiling as to the extent of economies of scale that can be achieved for manufactured homes. For another, manufactured home producers often have multiple facilities much closer together than the cost-effective driving distance. Legacy Housing has factories in Commerce, Texas, and Fort Worth, Texas, that are around 100 miles apart. Cavco has three factories in the Phoenix metro area, two factories in North Carolina that are 120 miles apart, and two factories in Georgia that are less than 75 miles apart. If there were substantial economies of scale at the factory level, we’d expect these nearby factories to be combined into larger operations.

Thus even in a homebuilding sector with many tailwinds enabling economies of scale, scaling effects still appear to be relatively minor. Manufactured housing is produced in a controlled factory environment and built to the standards of a single national code, both of which should enable the highly repetitive production necessary for capturing economies of scale. And yet the scale of most production facilities is low, factories employ little in the way of high-volume automated equipment, and very small-scale operators producing a few hundred homes a year appear roughly as competitive as companies producing tens of thousands of homes a year. These patterns suggest that there’s something about the homebuilding process itself that makes capturing economies of scale difficult.

What prevents economies of scale in homebuilding?

The easiest way to understand the difficulties of achieving economies of scale in homebuilding is to look at the relationship between input cost and output cost in various industries.

The graph below shows what’s known as the “crack spread” for gasoline production in the US Gulf Coast between 2014 and 2019. The crack spread is the difference between the price of crude oil and the wholesale price of gasoline. It essentially tells us how much margin oil refineries have to work with — how much they can afford to spend on the various production processes per barrel of oil processed. We can see that the crack spread for gasoline was typically around $15 a barrel for most of this period, compared to a crude oil price of around $50-70. In other words, the ratio of the price of wholesale gasoline to the price of crude oil was around 1.2 to 1.3.

An oil refinery is an enormous, expensive process facility that costs billions of dollars to construct. But because refineries produce such a large amount of gasoline and other petrochemicals, this cost can be spread very thinly across the refinery’s output, and the price of the final products is not all that much higher than the costs of the material inputs.

Likewise, consider car manufacturing. A few years ago, car industry analyst Munro & Associates conducted an analysis of the production costs of BMW’s i3 electric car. The report breaks down the various costs of production for different parts of the car, letting you see the cost of the inputs, the number and cost of the various operations performed on it, and the cost of the completed assembly.

We can see that the ratio of output costs to material input costs is higher than with gasoline production, around 1.8 on average. For complex manufactured goods that require a lot of distinct operations (the motor assembly alone has almost 5,000 parts and almost 30,000 assembly steps), it’s hard to achieve an output-to-input ratio as low as can be seen in a scaled continuous chemical production process.

More generally, any factory or production process will take some physical inputs, perform some set of transformations on them, and produce some series of outputs. If those transformations are involved or complex, requiring a lot of expensive machinery or labor-intensive operations, the ratio of the cost of the outputs to the cost of the inputs might be quite high. But even with a large number of complicated steps or very expensive equipment, this ratio can be driven down by taking advantage of economies of scale: spreading the fixed costs of your equipment out over a larger production volume, or using higher-capacity equipment or processes that has lower per-unit costs. As production volumes rise and a process gets more efficient, the ratio of the output costs to the input costs should theoretically approach one.

Elon Musk calls the ratio of the final cost of a product to the cost of its raw materials the “idiot index.” A high idiot index suggests a highly inefficient process with a lot of wasteful operations that could be dramatically improved. A low idiot index suggests a process with much less room for substantial improvements. (It was allegedly the high idiot index of rockets that convinced Musk he could beat existing aerospace companies and that SpaceX was worth pursuing.)4

Armed with this concept, let’s look at the cost breakdown of a typical US single-family home, courtesy of Craftsman’s National Construction Estimator. The chart below shows the hard costs of construction: what it takes to put up the actual, physical building.

The most important takeaway here is that the hard costs of constructing a single-family home are roughly 50% labor and 50% materials. This means that the ratio of the costs of the output to the costs of the physical inputs (drywall, lumber, concrete, etc.) is around 2. This is a very low ratio: it’s only slightly higher than the ratio we saw with various BMW i3 subassemblies.

With car manufacturing, this low ratio is achieved by way of scale, spreading the costs of expensive equipment and hundreds of workers over a very large output. If BMW tried to assemble its i3 cars one at a time, outside of a factory, the output-to-input ratio and the overall costs would be vastly higher, the same way that the craft method of car building was far more expensive than Ford’s mass production methods.

In home construction, however, this low ratio is a product of the fact that the amount of transformation taking place on the inputs is relatively modest. It takes a relatively small amount of labor and equipment to transform lumber, drywall, concrete, shingles, and other building materials into a finished home.

The fact that the ratio between costs of constructing a home and the costs of the various materials is already so low is fundamentally what makes achieving substantial economies of scale difficult. For complex manufactured goods, when this ratio is high — when there are a lot of labor-intensive operations or transformations required to turn input materials into a final product — the strategy for efficiency improvement is relatively straightforward. Build some machine to perform some particular task, sequence it together with other machines performing different tasks, and then use those machines to produce as much output as possible. The machines might be expensive to develop and build, and you’ll need new, expensive workers to keep the automated systems running. But with high production volumes this trade-off is worth it: all these costs get spread out thinly, and the ratio of output costs to input costs trends downward. This is broadly what Ford did for production of the Model T, and it’s how production for things like lightbulbs evolved in the early 20th century.

But when the ratio between output costs and input costs starts out low, as it does with homebuilding, it’s hard to take advantage of this strategy: the potential savings are just too small. I’ve noted previously that it takes a lot of expensive automation to duplicate the efforts of a small number of construction workers with hand tools, and when I was at Katerra, executives would often bemoan how hard it was for their capital-intensive production to compete with “Bubba and his truck” — low overhead contractors using manual labor and not much else. The ratio of output costs to input costs doesn’t give enough “bite” for production economies of scale to act on, and there’s little to be gained in production economies by dramatically increasing the scale of your homebuilding operation.

Conclusion

Economies of scale are often one of the most powerful forces available for making some process more efficient. But economies of scale operate on the difference between the costs of the raw inputs to some process and the costs of the final product. When this difference is large, there are often substantial economies of scale that can be achieved. You can introduce automated equipment or other large fixed costs that reduce your overhead costs when spread over a large enough production volume. You can use larger-capacity equipment that has lower per-unit costs than smaller-capacity equipment. You can gradually make some process more efficient via learning-by-doing. All these strategies will whittle away the difference between your raw input costs and the costs of your final product.

But when this ratio is already low, economies of scale have little room to operate. Thanks to the relatively modest requirements for transforming construction materials into finished houses, homebuilding has the sort of ratio of output costs to input costs observed in high-volume production operations like car manufacturing. Once the low-hanging fruit is captured, such as more efficiently arranging the work in manufactured home factories, there’s little production efficiency to be gained from further increases in scale.

Thinking about economies of scale in terms of the ratio of output costs to material input costs does suggest strategies for achieving them:

• You could find sectors of construction, such as nuclear power, where the ratio of output costs to input costs is very high, and thus have more opportunities for efficiency improvements.

• You could vertically integrate backwards into the production of raw materials and components, in the hopes of driving down those costs.

• You could find a way to use fewer and/or cheaper raw materials.

We’ll look at the viability of these strategies in future essays.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at squatter removal services, Apple finding uses for defective chips, process heat use in California, the brewing Colorado River crisis, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

War in Iran

Iran wants to charge US tech companies for use of the undersea cables that pass through the Strait of Hormuz. [Ars Technica] And Iran is starting a Bitcoin-backed ship insurance service [Bloomberg]

The hundreds of ships stuck in the Persian Gulf are starting to get damaged by all the barnacles and jellyfish that they’re accumulating. “...the Gulf’s shallow sandy seabed and warm waters have put ships at anchor or adrift there at risk of sand and sea creatures clogging up gratings that protect the vessel’s internal pipework. Seafarers are also struggling to get hold of critical parts when systems have broken down.” [FT]

An interesting example of the complexity of petroleum and petrochemical supply chains: a cooking oil shortage in India is driving up gas prices in California. “India, the most populous country, uses LPG as its primary cooking fuel. Cut off from Middle Eastern LPG, ​which represented over 90% of India’s total imports of the fuel before the Iran war, New Delhi has directed refiners to maximize LPG output. To comply, refiners have cut production of alkylates - motor fuel ​additives made using LPG as feedstock. For California, shrinking alkylate supply compounds concerns of a potential gasoline shortage due to declining fuel production and exports from ⁠Asian refiners struggling to access Middle Eastern crude oil. Alkylates are highly sought in California because they burn cleaner than other additives, and the state requires a unique gasoline blend to reduce smog.” [Reuters]

The US has lost 42 aircraft so far in the war with Iran, most of which (24) are MQ-9 Reaper drones. [National Security Journal]

Housing

The House passes its version of the Senate’s Road to Housing Act, without the build-to-rent restriction, 396 to 13. [NPR]

The NYT has a big, data-driven opinion piece on the need for the US to build more homes to bring down the price of housing. “The situation in expensive coastal areas, however, is far worse. They have enacted onerous zoning and building rules that limit home construction. They have allowed the “not in my backyard” instinct to prevail. Many of these areas vote Democratic and identify as politically progressive, yet their housing policies have increased inequality. By maximizing home prices, these parts of blue America have benefited existing homeowners, who tend to be older and richer, at the expense of everyone else. Nationwide, the relationship between home prices and home construction is even stronger than many Americans realize.”

Because California makes it so difficult to remove people living in a residence (whether they’re there legally or not), some California landlords are hiring “squatter removal services” to intimidate illegal squatters from vacating the properties they’re occupying. “Jacobs claims to have developed a long list of tools and tactics that enable him to remove squatters far faster than the court system, all while staying within the bounds of the law. Chief among them is a weapon he carries on every job: a katana, a curved Japanese sword that’s more synonymous with samurai warriors than clearing squatters. “In most industries, swords just don’t make any damn sense,” Jacobs says. “In this particular one, it actually does.” The lightly regulated katana, he explains, is an ideal weapon for indoor self-defense and intimidation.” [Reason]

Washington state legalizes scissor stairs, interlocking stairways that let you combine two stairways in one shaft. Scissor stairs are, like single-stair apartment buildings, a space-saving building design feature that’s common in other countries but mostly illegal in the US. [Sightline]

Manufacturing

Because of their microscopic size and their sensitivity to minute changes in chemical concentrations, semiconductor fabricators invest an enormous amount of effort into process control to minimize the number of defects that occur on a microchip. Defects nevertheless still occur, and when they do there’s often no real way to repair them. But that doesn’t mean the entire chip is useless; it just means you need to find a way to use it without the damaged areas. “The chip powering the Neo is Apple’s A18 Pro, the same chip first used inside the iPhone 16 Pro two years ago, but with one key difference. The Neo version of the chip has a “5-core” graphics processor, one less than the version inside the 2024 iPhones, indicating that Apple was able to save some of the A18 Pro chips with a defective core for future use. Defective cores can be disabled, leaving a chip that still functions perfectly well to power different, often cheaper devices—in this case an entry-level laptop instead of a top-of-the-line iPhone. It is the latest example of Apple deploying a decades-old chip industry strategy to squeeze profits from lesser-performing processors by selling them like eggs, gas, diamonds or hotel rooms, segmented by good, better and best.” [WSJ]

Read more

The build-to-rent, or BTR, industry has been in the spotlight recently because of a major federal housing bill, the Senate’s 21st Century ROAD to Housing Act. This bill, which is ostensibly designed to stimulate the building of new homes, includes a provision aimed at preventing large institutional investors from owning single-family homes. This provision, section 901, requires institutional investors (companies that own more than 350 single-family homes) to sell any build-to-rent homes to individual homeowners after seven years. Because BTR involves building a home and then retaining ownership of it to rent out, this provision threatens the fundamental business model of the BTR industry. Since the announcement of this provision, funding for new BTR projects has virtually ground to a halt while investors wait to see whether the bill actually passes. Over 100 pro-housing groups, including Berkeley’s Terner Center, the NAHB, and my colleagues at IFP have come out against this provision specifically, on the grounds that it’s likely to significantly reduce housing supply in the short term.

Because BTR has quickly become such a large fraction of new home construction and is now in the policy spotlight, it’s worth understanding the origins of the industry and why it has become so popular.

Origins of BTR

The modern BTR industry, where developers build entire communities consisting of dozens or hundreds of single-family homes for rent, is a product of the 2008 Global Financial Crisis. Prior to the financial crisis, single-family home rental wasn’t uncommon — in 2005, there were over 8 million detached single-family homes being rented — but the business was mostly the purview of small “mom and pop” operators that owned a relatively small number of scattered rental properties. As late as 2011, no single company owned more than 1,000 rental homes in the US.

But the financial crisis shifted the housing landscape. Huge numbers of people lost their homes to foreclosure: foreclosure rates in 2009 and 2010 were four times rates from 2005, and between 2007 and 2010, there were four million foreclosures. The homeownership rate in the US fell from a high of 69% in 2005 to 63% in 2016. At the same time, to rein in the subprime lending that had precipitated the crisis, banks tightened their lending standards, and average mortgage credit scores rose by more than 50 points. In 2003 buyers with a credit score of less than 620 made up 7% of all mortgages. By 2011 that had fallen to essentially zero.

The raft of foreclosures and the tightening of lending standards had two simultaneous effects on the housing market.

First, they pushed millions of Americans into renting. Between 2010 and 2015 the number of renter households in the US rose by roughly six million, while the number of homeowner households declined by roughly 800,000.

Second, this shift created a huge pool of homes available for purchase at very low prices. Between 2006 and 2010 the value of US homes dropped by 26%, greater than the average decline during the Great Depression. In some markets the declines were even worse: home prices declined by 60% in Las Vegas, and by roughly 50% in Phoenix, Miami, and Tampa.

In response to these market conditions — millions of homes available to buy cheaply, and millions of Americans who couldn’t afford to buy them — various real estate ventures were formed to take advantage of the situation. In 2010, the Arizona-based housing investment company Treehouse Group began to buy distressed mortgages in Phoenix and turn them into rental housing. Within a year the company had purchased 11,000 homes. In 2012 Treehouse was acquired by the investment group Blackstone, which turned Treehouse into the single-family rental company Invitation Homes. Today, Invitation Homes is one of the largest home rental companies in the US, with more than 86,000 rental homes across 12 states.

In 2012, the same year Treehouse was acquired, Wayne Hughes, founder of self-storage company Public Storage, founded American Homes 4 Rent, which similarly began to buy distressed mortgages and turn them into rental properties. Today, American Homes 4 Rent owns 61,000 rental homes across 24 states. 2012 was also when Tricon, a Canadian real estate company formed in 1988, began to buy up distressed mortgages for rental; by 2023, the company owned more than 38,000 of them. While prior to 2011 no single company owned more than 1,000 rental homes, by 2022 the four largest home rental companies owned more than 200,000 of them, largely concentrated in the Sun Belt and Western US.

This large-scale acquisition and transformation of single-family homes into rental properties was encouraged by the federal government, as part of broader efforts to keep the housing market from collapsing completely. In 2012 the Federal Housing Finance Agency launched the REO-to-Rental Initiative pilot program, which “allowed pre-qualified investors to bid on large portfolios of foreclosed properties owned by Fannie Mae.” Roughly 1,800 homes were sold to investors under this program. And in 2017, Fannie Mae backed a billion-dollar loan to Invitation Homes for the purposes of purchasing rental properties.

While most of the new entrants into the home-rental market focused on buying distressed properties, renovating them, and then renting them out, a few companies pursued a strategy of building rental homes directly. NexMetro, founded in Phoenix in 2012, completed its first community of single-family homes built specifically for rent in 2015, and today has built more than 10,000 rental homes across 50+ projects under its “Avilla Homes” brand. BB Living, another Phoenix-based company, also began building rental home communities in 2012, and today operates 18 of them. AHV Communities began construction on its first BTR project in Texas in 2014, and over the next 12 years would build three dozen more.

Building new rental homes had several advantages compared to acquiring existing homes. Being new construction, they typically had much lower maintenance costs than existing homes, and they could be designed by the developer with an eye towards minimizing maintenance and overheads. And because they were clustered together, they were somewhat easier to manage than purchased rental houses that might be spread across a wider area.

As the housing market recovered and the pool of single-family homes available for purchase at favorable prices dwindled, many of the large home rental companies began to experiment with their own BTR strategies. American Homes 4 Rent began work on its first ground-up rental community in 2016; today it owns more than 14,000 BTR homes, with essentially all new home acquisition coming through BTR. Invitation Homes began purchasing BTR homes in 2021 in a partnership with homebuilder Pulte, and as with American Homes 4 Rent essentially all its home acquisition now comes from BTR. Pretium Partners, which owns over 80,000 single-family homes under its “Progress Residential” umbrella, formed a $1 billion BTR venture in 2021. Some companies, such as American Homes 4 Rent, opted to do all their BTR development work in-house, while others preferred to partner with existing homebuilders, buying new houses that developers constructed in bulk.

BTR has attracted the attention of several large homebuilders. Lennar, the second largest homebuilder in the US, began building BTR communities in 2020 under its Quarterra subsidiary (though it has since sold a majority stake in this operation). DR Horton, the largest homebuilder in the US, began building BTR communities in 2019, and last year sold nearly 3,500 BTR homes. Taylor Morrison, the eighth largest homebuilder in the US, launched a BTR brand, “Yardly,” in 2022. BTR can be attractive for homebuilders because it provides them a large stream of steady work and can give them a sort of “escape hatch” if units seem to be selling to homeowners less well than anticipated. (Though apparently some homebuilders refuse to work with BTR companies because they don’t want to be associated with rental housing).

Today, BTR is still a small segment of the overall housing market: CBRE estimates that there are about 350,000 BTR units in the US, which is just 1.5% of the overall single-family home rental market. But it’s a rapidly growing segment of the US housing market — or was, until this recent Senate bill.

What sort of homes are BTR?

“Build-to-Rent” has become synonymous with single-family homes built specifically to be rented out, typically in communities of a few dozen to a few hundred rental homes, but within that category companies offer a broad range of different products. BTR generally gets broken down into several major subcategories (though some use slightly different ones): single-family detached, single-family attached, and horizontal multifamily. These categories exist on something of a spectrum of “very similar to conventional single-family homes” on one end and “very similar to conventional apartment buildings” on the other.

On the single-family home end of the spectrum, you have BTR communities such as Tricon’s Palomino Ranch, built in Houston in 2019. These BTR homes are essentially indistinguishable from single-family homes built for purchase: they’re detached homes with their own garages, driveways, and large fenced backyards. These sorts of homes will be “individually platted”: placed on their own individual plot of land with their own unique address.

On the apartment end of the spectrum, you have what’s known as “horizontal multifamily.” These are BTR developments that are essentially spread out apartment buildings. They are built in a range of sizes typically associated with apartment buildings rather than houses, with 1-, 2-, and 3-bedroom options, and they’re platted together on one large lot. They might have apartment-complex style amenities (gym, clubhouse, etc.), and will typically lack attached garages. They’re attractive to tenants because they don’t have shared walls and they have individual backyards, and they tend to rent for 15–20% more per square foot than a comparable apartment, but in many ways they’re closer to apartment complexes than they are to single-family homes. NexMetro’s “Avilla Homes Deer Valley” community in Phoenix is an example of this sort of BTR development.

Between these ends of the spectrum you have the “single-family attached” category — single-family homes that share one or more walls, such as townhomes or duplexes. BB Living’s Val Vista community of townhouses in Gilbert, Arizona is an example of this sort of project.

Regardless of the layout, a BTR community is at its core a large rental development and is designed by the builder to maximize rents and minimize overhead costs. For conventional homes built for sale, for instance, a builder might simply specify whatever finishes and features a buyer wants. But for a property that the developer is going to hold as a rental, choices are made with an eye towards minimizing maintenance costs by using cheaper, more durable finishes. American Homes 4 Rent, for instance, has stated that due to its designs being optimized for low-maintenance, maintenance costs are just 25% of what they were in their purchased homes.

Home rental companies also have other strategies for trying to minimize their overhead costs. Property taxes, for instance, are often the largest single operational expense for rental home companies. American Homes 4 Rent notes that it’s able to minimize its property tax costs by systematically filing property tax appeals to try and reduce the appraised value of its homes; a 2022 report noted that the company had a dedicated property tax team that filed more than 25,000 appeals annually.

Construction of BTR homes is in large part concentrated in Sun Belt states that have large populations, relatively high growth rates, and space to build new housing developments: Texas, Florida, and Arizona are the top three states for BTR construction, with Georgia and North Carolina rounding out the top five. An estimated 60% of BTR construction takes place in these five states. Within these states, BTR tends to be concentrated in large metro areas: most of Arizona’s BTR takes place in Phoenix, most of Georgia’s BTR is being constructed in Atlanta, and most of Texas’s BTR is being built in Dallas and Houston.

What’s driving BTR demand?

The largest factor driving demand for rental housing, both build-to-rent and buy-to-rent, seems to be housing affordability. Since 2012 home prices in the US have risen faster than inflation and median income. The post-Covid period in particular saw a dramatic increase in home prices, along with a large rise in interest rates that substantially raised the cost of mortgage payments and often made renting more financially appealing than buying. At the same time, Covid drove large increases in the frequency of working from home, which in turn created a demand for larger amounts of living space.

Americans who want more space or more privacy than apartments offer, but can’t afford to purchase a new home, will naturally turn to renting them. Amherst Group, which collectively owns over 50,000 single-family rentals (though not, as far as I can tell, BTR homes), noted in 2021 that 85% of its residents would not qualify for a mortgage due to low credit scores and incomes. NexMetro’s 2024 investor notes the largest factors driving demand for rental housing are rising interest rates and overall home affordability. American Homes 4 Rent’s 2025 annual report states that it has benefitted from “the surge in demand for larger living spaces, and increases in mortgage rates which have made home ownership more expensive.” One BTR developer I spoke to said that many residents simply don’t have the down payment that would be needed to afford a home purchase, and that BTR can give residents a way to access highly desirable school districts that would otherwise be totally unaffordable.

But while affordability issues seem to be the primary driver of BTR’s popularity, there also seems to be some fraction of residents that simply prefer renting over owning, due to a desire for less maintenance or simply because they don’t perceive owning a home as a major life goal. CBRE, NAHB, and NexMetro all mention various demographics of “renters by choice” (such as retirees), an analysis echoed by several BTR developers I talked to. As construction of BTR communities continues, this growth might create a sort of reinforcing cycle: more people move into rental housing, which makes it more accepted, which draws even more people in, and so on.

What’s BTR’s effect on the housing market and home prices?

The effect of large institutional investors (large companies that own thousands of homes) on the housing market has become a major point of controversy. Many, many people have the intuition that companies buying up or otherwise owning large numbers of single-family homes drives up the price of housing, and banning or curtailing this activity has become a popular cause on both the right and the left.

The intuition isn’t crazy. In a normal market, where producers aren’t limited in how much of a product they can make, this sort of purchasing wouldn’t drive up prices because it wouldn’t affect supply in a meaningful way. No one worries about Hertz driving up the cost of cars by buying huge numbers of them to rent out. But housing isn’t a normal market, and in many parts of the US it’s unreasonably difficult to create new housing supply. Ire at large, institutional homeowners would perhaps be more profitably directed at the various supply constraints that make it hard to build new housing in the first place (Invitation Homes notes that it benefits from operating in markets with “high barriers to entry”), but in a world where those supply constraints aren’t going to be removed any time soon, it’s worth considering what effect large rental companies are having on the market.

Supporters of rental housing are quick to point out that institutional owners make up a very small fraction of the housing market, less than 1% of single-family homes. However, because institutional ownership is highly concentrated in a few Sun Belt cities, in some housing markets institutional ownership can make up a substantial fraction (up to 25%) of the rental housing market.

There’s some evidence that institutional ownership might drive up home prices. Mills et al. (2015) noted that large scale buy-to-rent investors entering a market tended to “support” (i.e., prevent from falling) house prices in the short term. Lambie-Hanson et al. (2019) estimated that “the increasing presence of institutions in the housing market explains over half of the increase in real house price appreciation rates between 2006 and 2014.” D’Lima and Schultz (2020), Oosthuizen (2023), Coven (2025), Wang and Zhai (2026), Barbieri and Dobbels (2026) and Gorback et al. (2024) all found broadly similar results. However, Hanson (2024) finds that while house prices rise when institutional investors enter a market, this rise can mostly be attributed to improving market conditions, and that prices would have increased whether institutional investors entered the market or not.

The evidence for the effects of institutional ownership on rents, however, is somewhat different. Large owners might use their market power to raise prices (large rental companies will basically admit they raise rents compared to what mom and pop renters would charge), but by buying properties and turning them into rentals investors are also adding supply to the rental housing market, which all else being equal should bring rents down. Lee and Wylie (2024) find that investors raise rents at above average rates when first entering a market. Gurun et al. (2023) found that “institutional landlords leverage their market power to extract greater surplus from renters.” And Hanson (2024) finds that rent growth is due to the market power of the institutional investors. On the other hand, Coven’s (2025) model based on various census and investor property data suggests that added supply from institutional investors can reduce the rent paid by renters, and Wang and Zhai (2026) find that “the majority of renters gain from the expanded rental supply.” Barbieri and Dobbels (2026) find that while the market power of institutional owners raises rents, the increase in supply more than makes up for it, causing rents to fall on net. Economics student Nicholas Decker looks at what he thinks are the strongest papers, and concludes that the net effect of institutional investors entering a market is positive for consumers overall, with the benefits of lowered rents outweighing the increase in house prices.

However, these studies are all for buy-to-rent homeowners: investors buying up homes that already exist. None of them looked at build-to-rent, which we might expect to have substantially different effects: buy-to-rent reallocates existing supply, whereas build-to-rent creates new supply (though it could theoretically also reduce the supply of homes for sale if, in the absence of BTR, for-sale housing projects would be built in the same location).

Unfortunately, the economic research around the effect of BTR specifically is much thinner, and there haven’t been robust studies of the effect of BTR on the housing market specifically.

Conclusion

Overall, I think there are a few points worth making about BTR.

One is that, given that the attractiveness of owning vs. renting can vary depending on the person and the state of the housing market, making a broader array of rental options available for people is a positive thing. I remember years ago when I was trying to find a large (3+ bedroom) place to rent, and how difficult it was to find rentals that size in the area I was looking in. Giving people more ways to purchase housing is good, the same way it’s good for people to have the option to buy or lease a new car.

Second, it’s clear that many folks strongly believe that large-scale corporate ownership of rental housing (which would include BTR communities) is something that can have negative effects on the housing market. But I think it’s more useful to think of the popularity of rental housing as something that’s a product of the housing market: it’s a natural consequence of housing getting increasingly unaffordable thanks to high interest rates and skyrocketing housing prices. Shutting down BTR is a poor way to address that problem; what we need to do is build more housing and develop construction methods that let us construct buildings more cheaply.

Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure and industrial technology. This week we look at Tokyo’s cheap housing and expensive land, the House response to the Senate housing bill, an IED near an Alabama dam, Fervo’s IPO, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.

Housekeeping items:

• No essay this week because of a respiratory illness that is carving through our household like Sherman through Georgia, but I’m working on a piece about the build to rent housing industry that should be out next week.

• I gave a talk about why major building collapse is so uncommon at Antithesis’ Bug Bash a few weeks ago. You can watch the talk here.

War in Iran

The Economist on why the closure of the Strait of Hormuz doesn’t seem to have yet created a worldwide oil crisis: other exporters (like the US) have raised their output, and many larger importers (like China) are importing less. [The Economist]

On the other hand, it seems like supplies of various petroleum products are starting to get thin. The owner of an oil change company is claiming that US consumer motor oil supplies are going to start getting thin in a few weeks. [X] And European jet fuel stockpiles are apparently almost exhausted. [X]

Some Japanese snack manufacturers are changing to black and white packaging due to ink supply disruptions. [AP News] The disruptions are due to a shortage of naphtha, a light fraction of crude oil. Interrupted naphtha supplies are also apparently poised to cause disruptions to the Japanese semiconductor industry. [Threads]

The UAE has started to build large anti-drone cages around its infrastructure to protect it from Iranian drone attacks. [The War Zone] And it’s also building a second oil pipeline to increase how much oil it can export outside the Strait. [X]

Thanks to Trump’s Jones Act Waiver, there have now been at least 45 cargo trips between US ports via non-US ships. [Cato]

Housing

Alex Armlovich on how making it easier to build housing both reduces rents and increases land values in the city of Tokyo. “The famous (relative) cheapness of Tokyo housing is not a story about cheap land. A one-acre detached house in central Tokyo would cost more than $100 million in dirt before you broke ground. The land is pricey but the structures are cheap. Admittedly, Tokyo’s rents and prices are not as cheap per square foot as buildings in the US Sunbelt’s midsize cities, but cheap by the standards of any 10-million-plus Anglosphere metro area. Tokyo built its way to relative affordability without ending up with low land values, and the values themselves look reasonable for a productive, agglomerated megacity that simply didn’t artificially restrict its own supply.” [Abundance and Growth]

Ezra Klein moderates a debate on housing policy between the leading Democrat candidates for Governor. [YouTube] And CA Yimby declined to endorse any candidate for governor, not because none of them had good housing plans, but because all of them did. [Cal Matters]

The house comes out with its response to the Senate’s 21st Century ROAD to Housing Act. The house bill notably does not contain the onerous restrictions on build-to-rent housing that the Senate bill does. [Politico]

Are homes sitting empty because it’s too expensive for people to sell them? “According to Flock Homes, for many retirees “the tax bill triggered by a sale far exceeds the cost of simply leaving the home empty.” This includes capital gains taxes (paid on the profit from the sale) and depreciation recapture taxes (which sellers of rental properties face). The federal capital gains exemption, which applies to the sale of primary residences, was not factored into the analysis of vacant homes. Mortgage payments were not included in the analysis, since many older homeowners own their homes outright.” [NYT]

Number of sellers vs number of buyers in the US housing market. [X]

Manufacturing

Read more