Which City Builds Skyscrapers the Fastest?
Last week we looked at trends in skyscraper construction speed for New York and Chicago, finding that New York has gotten significantly slower at building skyscrapers over time. Chicago, on the other hand, has declined in speed less steadily, and currently builds skyscrapers much more quickly than New York does.
The obvious next step is to look at skyscraper construction speed around the world. How do modern New York and Chicago compare with other cities, both in the US and abroad? Which cities build skyscrapers the fastest and the slowest today? Let’s take a look.
I started with Wikipedia’s list of 50 largest cities in the world. For each city, I pulled data from the Council on Tall Building and Urban Habitat (CTBUH)'s skyscraper database on every skyscraper completed between 2000-2020 that was taller than 100 meters, had a start and completion date, and had a gross floor area. For most cities this was a small fraction of the total number of skyscrapers in the dataset, so many buildings aren’t included in this dataset.
Some large cities turned out to not have enough data: either they didn’t build enough skyscrapers, or the skyscrapers they did build didn't have the necessary information, or both. For instance, Paris only has a single skyscraper completed in that time period in the database, the Tribunal de Paris building. And because gathering information for the Chinese cities was so time consuming, I stopped after the first six (the five largest plus Hong Kong).  And I added several other (mostly American) cities that weren’t in the top 50 but nonetheless seemed interesting. The result was a list of 986 skyscrapers completed in 39 cities around the world. Most of these (~740 of them) were in just four countries: the US (261), China (242), Japan (152), and Canada (89).
The chart below shows average construction speed, in square feet per year, for cities in the US, Canada, China, and Japan:
And here’s cities in the rest of the world:
I did not expect these results. Although I thought the US would be comparatively slow relative to China and Southeast Asia, average speeds in the US and China are actually fairly close (an average of 294k vs 311k square feet per year, respectively, across all skyscrapers). Within each country we see a range, with some slow cities (New York, Hong Kong), and some fast (Beijing, Chicago). New York remains a particularly slow city, but it's not an enormous outlier (which is what I expected): Singapore, Toronto, and Hong Kong are all slower. Not only does Chicago build skyscrapers faster than New York, it builds them faster than most other cities around the world (in this dataset, at least).
We also see that Japan is significantly faster than the US on average, with an average speed of 379k square feet per year. But the fastest US cities do roughly as well as the fastest Japanese ones. The fastest city out of any examined is, somehow, Los Angeles, though random variation may be the cause (it only has five skyscrapers in the database). Interestingly enough, the huge outlier in slow construction isn't the US, but Canada, with an average skyscraper construction speed of half that of the US’s.
If we look only at the very tallest buildings, China builds much faster than the US. Of the five tallest buildings built in the US and China in the past 20 years, the Chinese ones were built 80% faster than the US ones (partly driven by the US list containing two comparatively slow pencil skyscrapers). But the trend doesn’t hold as we look down the list. For the top 25 in the dataset , Chinese skyscrapers are only 40% faster on average. For the top 50, they’re only 20% faster. The largest skyscrapers in China took four to six years, or longer, to construct. Shanghai Tower took six years to build, as did CITIC Tower. Ping An Financial Center took seven. China is constructing a huge number of huge buildings, but they don't seem to be going up in a matter of months.
A moving average of construction speed over time doesn't change this overall picture. Japan is fast (and getting faster), Canada is slow (and getting slower), and the US and China are staying fairly steady in between.
Why did I expect the US to be slow and China and Southeast Asia to be fast? I may have confounded micro and macro speed. At a macro level, many Southeast Asian cities have been built incredibly quickly. Over 30 years, Shenzhen went from a population of 880,000 to more than 11 million:
Plenty of other Southeast Asian cities have similar transformations. Here’s Singapore:
And here’s Hong Kong in 1967 and 2021:
Because these cities get large so quickly at a macro level, I assumed that the buildings themselves would get constructed incredibly quickly. This impression was probably encouraged by the occasional video of something getting built in China incredibly quickly.
Conversely, one of the most salient facts about US construction is how sluggish it is. Everything from nuclear reactors to apartment buildings to transit projects gets built more slowly than it used to, federal projects get slowed down by years-long NEPA reviews, housing construction is at historically low rates, etc. I assumed the factors slowing down US construction would be unique to the US (or at least the west), and wouldn’t be seen in rapidly growing Asian cities. But none of the cities examined, not even the Chinese ones, build skyscrapers as fast as New York City did from the 1930s through the 1960s. Only two skyscrapers in the list were built more than half as fast as the Empire State Building, and neither of them were in China. Not only is the US building more slowly than it used to, but (apparently) so is everyone else.
It is, of course, possible this is all just a data artifact. The Chinese cities, in particular, had a comparatively small fraction of buildings with completion information, particularly for the repetitive concrete highrises that are so common there. And I deliberately omitted some Chinese cities in the top 50 in the interests of time. So these construction rates might look different if more data was available.
But it would surprise me if the limited dataset creates enormous distortions. If you had the ability to build skyscrapers much faster, why wouldn’t you use it on your tallest, most expensive buildings, where it would be most valuable and impressive? Also, some Southeast Asian cities (notably Tokyo) have completion information for a large fraction of the buildings in the database, and their construction speed is consistent with what we see in other cities. And looking for videos of extremely fast skyscraper construction in China doesn’t turn up a huge number of examples: the same two or three examples built by Broad Sustainable Building repeat over and over again, and those used an atypical skyscraper construction system that (as far as I know) is no longer used.
Floor plate size?
I previously noted that a difference in floor plate size could possibly explain the difference in construction speed between New York and Chicago skyscrapers, with smaller floor plates taking proportionately longer to build than larger ones. Unfortunately, I realized that most of my previous analysis is incorrect due to the presence of autocorrelation. Briefly: autocorrelation is a spurious correlation, driven by comparing a dataset against itself. For instance, if we graph construction speed vs floor plate size from this dataset, we get what looks like a strong correlation, suggesting smaller floor plates get built slower:
However, what this graph is actually graphing is (building area / stories) against (building area / time): that is, the apparent correlation comes from comparing building area to itself, not because there’s necessarily a relationship between floor plate size and construction speed. I can get a similarly large correlation with completely random story and time data:
Unfortunately, this sort of autocorrelation is lurking within my previous graphs of floor plate size vs construction speed (as well as building size vs construction speed). When breaking construction speed into size buckets, I'm comparing building area, or (building area / stories), against (building area / time), again comparing building area with itself. Again, I can use completely random story and time data to make it look like smaller floor plates build more slowly:
What happens if we look at the effect of floor plate size on construction speed while avoiding problems of autocorrelation?
To start, we can compare building size with construction speed measured in years, rather than square feet per year. This will remove the “building area” variable from the y axis and the resulting autocorrelation. Let’s first look at correlation between number of stories and construction speed in years.
Taller buildings take slightly longer to build than shorter ones, but the correlation is pretty weak. We see a similarly weak correlation if we look at total floor area vs years.
And floor plate vs years:
Looking at individual cities gives basically the same picture: a relatively weak correlation. Here’s Tokyo:
And here’s Shanghai:
And here’s Chicago:
Looking at more cities, it’s mostly the same story, though we see a few stronger correlations here and there. But in general, building speed (in years) and building size (in either stories or floor area) are not particularly related. Larger buildings don’t take much longer to build than smaller buildings. And rather than being a good predictor of construction speed, floor plate size actually seems particularly unrelated:
If we keep the number of building stories constant, we see something similar. For a given number of stories, larger buildings aren’t built any slower than small buildings:
Likewise, if we keep building area (roughly) constant, we see that taller buildings aren’t built any slower than shorter buildings:
So not only does floor plate size not correlate with construction speed, but building size in general seems to have a surprisingly small impact on construction speed. Although we might expect large skyscrapers to be built much more slowly than smaller ones, it seems a modern skyscraper takes between 2-10 years to construct, which varies depending on the city, regardless of how big it is. If you build bigger buildings than average, it’ll look like you’re building faster on average, and if you build smaller buildings it’ll look like you’re building slower, but really speed and size are only weakly related (though some cities show stronger correlations than others).
 - Cities omitted because of lack of data: Delhi, Cairo, Dhaka, Karachi, Lagos, Rio de Janeiro, Kinshasa, Lahore, Bangalore, Paris, Bogota, Kolkata, Chennai, Hyderabad, Tehran, Madrid, Berlin, Munich, Rome, Phoenix, Dallas, San Diego
Chinese cities in the top 50 I skipped: Changdu, Nanjing, Wuhan, Xi'an, Dongguan, Hangzhou
 - Not necessarily the top 25 overall, since this is a limited set of cities.