AI data-center capex to hit 3.1% of GDP, twice housing pace

Hyperscaler data-center capital spending is projected to reach 3.1 percent of U.S. GDP by 2027, building at nearly twice the pace of the 2000s housing boom and raising the risk of an abrupt reversal.

"A cycle that builds at 0.85 percentage points a year can unwind at a similar pace, and that, rather than the buildout itself, is the macro risk if AI demand disappoints," said Torsten Slok, partner and chief economist at Apollo Global Management.

Data-center capex rose from 0.3 percent of GDP in 2019 to 1.4 percent in 2025, and consensus forecasts put it at roughly 3 percent annually from 2027 through 2029. The two-year jump of 1.7 percentage points works out to about 0.85 percentage points a year, nearly double the 0.5-point annual pace of the housing boom's fastest phase from 2002 to 2005. The telecom and fiber buildout of the late 1990s expanded at just 0.15 percentage points a year and peaked at 1.2 percent of GDP in 2000, while residential investment topped out at 6.6 percent in 2005.

The stakes are large because the U.S. economy produces roughly $28 trillion a year, making 3 percent of GDP about $840 billion directed at a single category of infrastructure. Five of the biggest U.S. technology companies — Amazon, Alphabet, Meta Platforms, Microsoft and Oracle — are expected to spend about $800 billion in combined capital expenditures this year, and the four largest hyperscalers have committed nearly $2.4 trillion over the next few years.

The cumulative change is what matters for GDP growth. Data-center capex is set to climb 2.5 percentage points of GDP from 0.6 percent in 2023 to 3.1 percent in 2027, a bigger swing than either the telecom cycle's 0.4-point rise or the housing boom's 2.2-point increase from the mid-1990s to 2005. "The bottom line is that the data-center buildout is smaller than housing in level but larger in the change in share of GDP, and faster than either previous cycle," Slok wrote.

The warning is rooted in the arithmetic of reversals. Housing investment's collapse from 6.2 percent of GDP in early 2006 to 3.0 percent by the end of 2008 was the primary driver of the severity of the Global Financial Crisis recession, while the telecom bust's smaller reversal produced only the mildest postwar downturn. Slok's concern is that a cycle building at its current speed could contract with equal force if the underlying demand for AI services fails to materialize.

The analysis has drawn attention from investors who called the last bubble. Michael Burry, who bet against U.S. housing before 2008 and has since taken short positions in Nvidia and Palantir, shared Slok's charts on social media. Steve Eisman, another investor profiled in "The Big Short," pushed back on the timing, saying the break might still be "a year from now" and that he needs a price war among large language model developers — OpenAI, Anthropic and cheaper Chinese models — to confirm the unwind. Roughly 70 percent of Microsoft's AI revenue reportedly came from OpenAI last year, according to a Bloomberg analysis.

Prediction markets lean toward Eisman's view on timing. On Polymarket, a contract defining an AI bubble burst requires at least three qualifying events within 90 days, including Nvidia falling 50 percent from its all-time high, an OpenAI or Anthropic bankruptcy, and H100 rental prices at $1 or lower for five straight days. Traders price a 15 percent chance of that scenario by Dec. 31, with $2.9 million in total volume wagered.

The debate arrives as the AI trade extends its run. The Invesco QQQ Trust is up 26 percent over the past 12 months, while the Global X Artificial Intelligence & Technology ETF has surged 39 percent. Nvidia is scheduled to report second-quarter results on Aug. 26 after guiding for roughly $91 billion in revenue. If the capex plateau arrives and revenue does not follow, the pressure would concentrate in the suppliers — chipmakers, power companies, data-center real estate investment trusts and smaller software firms — rather than at the top of the stack.

This article is for informational purposes only and does not constitute investment advice.