AI capex boom is growing twice as fast as the housing boom did

The last time the U.S. economy saw capital pour into a single sector this fast, it ended with a financial crisis. That comparison is not a prediction. It is, however, the framing that Torsten Slok, Partner and Chief Economist at Apollo Global Management, is putting on the current AI infrastructure buildout.

Slok’s analysis tracks hyperscaler data-center capital expenditures as a share of U.S. GDP. The numbers move quickly: from 0.3% of GDP in 2019, to 1.4% in 2025, and a projected 3.1% by 2027. That two-year jump of 1.7 percentage points works out to roughly 0.85 percentage points per year.

Why the housing comparison matters

During the housing boom from 2002 to 2005, residential construction added about 0.5 percentage points to its GDP share annually. The AI capex buildout is running at nearly double that pace.

To be clear about scale: the housing market’s contribution to GDP peaked at 6.6% in 2005. Data-center capex, even at its 2027 projection, sits at less than half that level.

There is another historical data point worth anchoring to. The telecom boom of the late 1990s peaked at around 1.2% of GDP in 2000. The AI buildout is on track to more than double that figure, and it is doing so with far more established revenue bases behind it. Amazon, Meta, Microsoft, Alphabet, and Oracle are not speculative startups; they are profitable businesses making enormous forward bets on infrastructure they believe they will need.

What is actually being built

The projected spending trajectory suggests this does not slow down after 2027. Apollo’s analysis points to data-center capex holding around 3% of U.S. GDP annually from 2027 through 2029. That is a sustained plateau at a level no technology infrastructure sector has previously maintained.

For context on what 3% of U.S. GDP actually represents: the U.S. economy produces roughly $28 trillion per year. Three percent of that is in the neighborhood of $840 billion annually, directed almost entirely at a single category of technology infrastructure.

The risk Slok is flagging

Apollo’s chief economist is not arguing the AI boom will collapse. His warning is more precise than that: rapid build cycles tend to unwind rapidly. Companies spend ahead of demand based on their best guess of what they will need. If the demand materializes, the capex looks prescient. If it does not, the assets sit underutilized, and the companies that built them pull back hard and fast.

AI infrastructure does not carry the same debt-financing architecture as residential housing. There are no mortgage-backed securities tied to GPU clusters. But the macroeconomic arithmetic of a sudden pullback in spending worth several percentage points of GDP would still be significant.

The more exposed players are the suppliers: chip manufacturers, power companies, data-center real estate investment trusts, and the smaller software companies betting that hyperscaler infrastructure will translate into enterprise AI demand for their products. If the capex plateau arrives and revenue does not follow, the pressure will concentrate in those layers of the ecosystem rather than at the top of the stack.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.