New bubble, same old anxiety
Apollo chief economist Torsten Slok just threw a big, loud “hey, maybe don’t ignore this” onto the AI trade. His latest take: data-center investment is ramping up so fast that it’s closing in on the speed of the mid-2000s housing boom — the same boom that later helped turn a slowdown into a full-blown recession.
The headline number is a little wild. Slok’s analysis says data-center capex could climb from 1.4% of GDP in 2025 to 3.1% in 2027, which works out to about 0.85 percentage points a year. For comparison, housing at its fastest only ran at about 0.5 points a year, and telecom/fiber in the late ’90s was a sleepy 0.15.
The real villain: the unwind
Here’s the part that matters for markets: the risk isn’t just that AI spending is huge. It’s that it could also fall fast if demand disappoints.
That’s the nasty little replay investors should keep in the back of their heads:
- Housing didn’t just hurt because it got big; it hurt because it rolled over hard.
- Slok argues the same logic could apply to AI infrastructure if hyperscalers slam the brakes.
- If capex unwind happens at roughly the same speed it built, the macro hit could get ugly pretty quickly.
Why stocks care even if this isn’t a “company story”
This isn’t about one ticker. It’s about the market’s favorite growth narrative becoming a potential macro plot twist. When a theme gets this big, it stops being just a tech story and starts acting like a GDP story.
That’s why traders watching the S&P 500, Nasdaq, and Dow should care: if AI capex keeps juicing growth, great. But if the trade reverses, the same spending that powered the rally could become the thing that drags on it.
Big picture: AI may still be in its “move fast and build things” era, but Wall Street is starting to ask the awkward question — what happens if the buildout becomes the bust?