The AI buildout is 3.6% of GDP a year — more than any US boom
A new Brookings paper prices the US AI buildout at $10.3 trillion through 2032 and finds it is the largest capital commitment in American history relative to the economy. Separately, an engineer left Google rather than help make AI cheaper.
The US AI buildout is projected to consume $10.3 trillion between 2025 and 2032 — about 3.63% of GDP every year for eight years, a share larger than any previous American infrastructure boom, including canals, railroads and electrification combined. The figure comes from a Brookings paper by Columbia economist Stijn Van Nieuwerburgh, presented today at the Brookings Papers on Economic Activity conference. The comparison that makes it land is the per-year share of GDP, not the headline total: canals ran 0.66% of GDP a year from 1836 to 1841, railroads 2.24% from 1870 to 1890, the electricity grid 0.50% from 1905 to 1925, the highways 1.13% from 1956 to 1973 and the telecom buildout 1.10% from 1996 to 2003. Add canals, railroads and the grid together and you get 3.40% — still below AI's 3.63%.
Two framing notes worth keeping, because the round-number version of this story is already circulating. The $10.3 trillion covers data centers plus the power systems, networks, chips and related equipment around them, not server halls alone. And the paper's own conclusion is deliberately unalarming: "It would be premature to conclude that AI infrastructure already poses systemic risk," the authors write, arguing that the most useful policy contribution right now "may be to improve measurement and transparency" — not to slow anything down.
The supporting numbers point the same direction. Goldman Sachs puts US AI investment at 1.9% of GDP this year, the first time a single new industry has taken that much of the economy since the late-19th-century railroads. Commerce Department data show $37 billion of private data center construction through July 2026, up $9 billion year over year — while all other private construction fell by $46 billion. Alphabet, Amazon, Meta, Microsoft and Oracle have committed roughly $4.2 trillion of capital spending through 2029, and a growing share of it is debt-financed, which is where the systemic-risk question actually lives.
We have covered the revenue side of this math before — the hyperscalers' trillion-dollar buildout and whether earnings can grow into it — and this paper's contribution is the denominator. Every previous American buildout was financed against a known revenue model. This one is being financed against a bet that models keep getting more useful, which is precisely why "improve measurement and transparency" is a bigger ask than it sounds.
A Google engineer resigned over his own team's work, because he could not pretend the AI impact was ambiguous. Robert O'Callahan — Mozilla Distinguished Engineer, author of the record-and-replay debugger rr, founder of Pernosco — published his resignation email and a short essay explaining it. He was not working on AI capability; he was building improved tools for hardware chip design in Google's New Zealand office. Over time he concluded the main effect of those tools would be to accelerate a new breed of AI chips, making AI "much cheaper and lower-latency." He told his skip manager he was "at best reluctant to see their project succeed," found no non-AI team to move to, and left after roughly four and a half years.
The part that makes it more than a personnel story is the honesty about the mechanism. O'Callahan writes that he wishes AI would "hit some kind of plateau" and does not expect it to, and that the rate of change — not necessarily the destination — is what he objects to. He is explicit that this was not a sponsored exit: the date pre-dated a ten-day tramp, and he is staying in the field, maintaining rr and Pernosco and working on how AI agents debug code. A front-page Hacker News thread (109 points, 76 comments) took the argument seriously in both directions.
The wider context is that departures like this are now frequent enough to read as a pattern rather than a signal. The researchers who left say AI transparency is voluntary — Trump says he has no extinction concerns tracked the earlier wave out of frontier labs, and those exits came from safety teams. This one came from the chip-tooling side, from someone who never claimed to work on AI at all — which is the more uncomfortable version of the argument.
What to watch: whether anyone with budget authority answers Van Nieuwerburgh's transparency recommendation, and whether mainstream tech press picks up the O'Callahan post or leaves it a community-only item.
The buildout is now financed on the assumption that models keep getting more useful — is that an investment thesis or a bet on continued improvement? Tell us in the comments.
Sources: Brookings — Financing the AI buildout · The Wall Street Journal — The AI build-out is becoming the biggest economic bet in US history · Robert O'Callahan — Goodbye Google · Hacker News discussion · Lobsters discussion
