The AI boom has a borrowing problem.
For two years, the question was how much AI demand there was. Now it’s who can pay the interest. Investors are worried that AI revenue won’t be enough to service the debt taken on to build AI infrastructure, and credit default swap prices and yields on AI-related bonds have risen sharply.
The poster child. CoreWeave, a crypto miner four years ago, is now the best-known of the “neocloud” companies. Carrying about $50 billion in debt, its CDS spreads have hit about 1,000 basis points. That implies a nearly 63% chance of default within five years, up from about 30% a year ago.
It isn’t only the small players. Oracle’s five-year CDS spread widened to a record 288 bps, SoftBank’s reached 424 bps, and spreads for Amazon, Alphabet, Meta and Nvidia also hit record wide levels, though at much smaller spreads than Oracle.
Junk money is doing the building. Of the $80 billion in data-centre bonds raised so far this year, $37 billion, almost half, is junk-rated. That’s more than four times the 2025 total. Bonds already trade below face value. Debt for Meta’s Hyperion project has slipped to just under 92 cents on the dollar, while Blackstone’s Microsoft-leased QTS site in Georgia is below 88 cents.
The maths gap. Gartner estimates AI data-centre owners need $8.5 trillion in revenue by end-2030 to cover their capital and financing costs and earn a healthy return, against the $3.6 trillion it expects customers to actually spend. That’s a gap of roughly 2.4 times. Gartner also expects over 80% of neocloud companies to stop operating independently by 2030.
The physical world is pushing back. At least 47 large projects have been delayed, withdrawn or blocked in 2026, and more than 1,000 US communities have moved to curb AI infrastructure this year.
Why it matters. Allianz’s CIO called this a “Darwin moment” for data-centre bonds. With three hyperscaler earnings reports and a Fed decision due next week, the AI trade now depends as much on the bond market as on demand for chips.