In the latest episode of Power and Consequences, we (Gary Gensler and Simon Johnson) discuss the AI data centers building boom and its likely consequences. Is the current pace and extent of data center construction an essential part of innovation, economic growth, and competition, or are we overbuilding in the U.S. (and perhaps elsewhere)? This is first of a two part series — next week we’ll discuss should we (or can we) slow the development of AI?
The growing investment in data centers is eye-popping. Amazon, Microsoft, Google and other US hyperscalers’ capital expenditures has grown nearly five-fold over the last three years to about $750 billion (projected) in 2026. Analyst consensus estimates are for this to grow in 2027 to as much as $1 trillion. With 2700 data centers already in operation (Data Center Map), the US currently is building 400 and planning another 1500.
Are expectations for revenues from the use of AI (and the prices that can be charged) overblown? How will China’s open-weight AI models affect the competition for future revenues? Though the US spends nearly seven times more for data centers annually than China, the US model performance lead over China is only measured in months.
Economic history contains many past building booms with compelling underlying rationales alongside excessive exuberance — think about what happened with canals, railroads, electricity generation, and the early Internet.
What might these earlier booms foretell about our economic future? No building boom lasts forever. When the slowdown comes, what will be the consequences? A major part of the answer will lie in the amount, complexity, and precise holdings of debt related to data centers and their suppliers. Adding to these risks, there’s a great deal of opacity around these arrangements. Could we be heading for a serious case of economy-wide and financial markets indigestion?
There also are many micro concerns about data centers, focused on the implications for local communities. What are the effects on electricity prices, water use, the environment, and noise? Data centers are being built faster than electricity capacity can be added. The cloud providers, AI model labs, and the White House seem to have consistently underestimated local opposition. While some jobs are created, not many remain once data centers are completed — and there are broader concerns about what AI will do to existing human jobs of all kinds. Political pushback is likely to intensify.
Please join us again next week when we’ll take on the other big question of the moment: Should we (or can we) slow the development of AI?
