Apollo's Torsten Slok Is Comparing AI Debt to 2008 Housing. He's Not Wrong to Try.

I've covered enough bubbles to know that the 2008 comparison gets trotted out roughly every time a credit market sneezes. So when Apollo Global Management's chief economist put hyperscaler debt next to the housing collapse on Wednesday, my first instinct was to roll my eyes.

Then I read the note. Torsten Slok isn't yelling fire in a crowded theater. He's pointing at a spread, and the spread is moving.

The Numbers Behind the Warning

Here's the mechanical part. A credit default swap is basically insurance on a company's debt. When it gets more expensive, someone is paying up to protect against the borrower stumbling. Slok flagged that CDS on hyperscaler debt has widened to about 60 basis points against bank debt, and he tied that move to weakening credit fundamentals rather than the usual bond-dealer hedging noise. That distinction matters more than it sounds like it should.

Granted, dealers hedge all the time. Admittedly, that creates plenty of false alarms. But Slok's argument is that this one isn't noise. It's the market repricing the credit quality of the companies building the AI boom.

The reason is simple enough. Hyperscalers used to fund their data centers out of pocket. Amazon, Microsoft, Alphabet and Meta were cash machines, and capex came straight from the till. That's changing. Meta sold $30 billion in bonds back in October, the largest corporate debt deal of the year. Oracle borrowed $18 billion. Combined hyperscaler capex is running north of $300 billion for 2025, and a growing slice of it's financed with other people's money.

So the AI trade now has a bond-market appendix.

Why This Reaches Past the Bond Desk

If CDS spreads keep widening, the cost of funding the next data center goes up. That sounds abstract until you follow it downstream. Higher funding costs make marginal projects pencil out worse. Fewer marginal projects means softer orders for GPUs, slower leasing for data center operators, and less demand for the power contracts everyone's been signing at a frantic pace.

And crypto isn't a spectator here. Bitcoin still trades like a liquidity asset, and it tends to feel it first when credit conditions tighten. The miners who pivoted to hosting AI workloads are directly exposed to hyperscaler budgets. So are tokenized credit products and the stablecoin reserves that hold corporate paper. This isn't some distant Wall Street problem we get to read about on a screen.

Color Me Skeptical, But

I'm not entirely convinced the 2008 comparison holds up. Housing was subprime mortgages, stuffed into opaque securities, held by thinly capitalized firms, and marked to market every single day. Hyperscaler debt is investment grade paper from companies with cloud and advertising revenue that could cover the interest several times over. That's not the same animal. History suggests otherwise when people try to force the parallel.

But the reflexive loop is real. The AI narrative is what keeps the credit cheap. The cheap credit is what funds the capex. The capex is what keeps the narrative alive. Break one link and the others wobble.

The question worth asking: what happens to AI capex if the cost of financing it doubles?

Nobody knows. That's the honest answer.

What I'd watch next is boring and specific. New issue concessions on the next big hyperscaler bond deal. Whether rating agencies start moving outlooks. And whether that 60 basis point gap keeps widening or settles back toward normal. If it settles, Slok's note becomes a footnote. If it doesn't, the skeptics get a much louder voice.

Either way, the AI trade just picked up a credit market referee. That part is new.