The AI infrastructure boom just hit a wall made of debt, and CoreWeave's 15% bond yields are the canary in the coal mine.
The Summary
- US chip stocks fell for 8 straight days, dropping 16% in their longest losing streak since the 2008 financial crisis
- Rising borrowing costs are testing AI industry financing, exposing the fragility of debt-fueled infrastructure buildouts
- CoreWeave's bonds hit nearly 15% yields as traders buy bubble puts, signaling serious credit concerns in the neocloud sector
- The selloff reveals a fundamental tension: AI infrastructure requires massive upfront capital, but the revenue models haven't proven they can service the debt at today's rates
The Signal
The semiconductor index just posted its worst streak since Lehman Brothers collapsed. Chip stocks fell 8 consecutive days, wiping out 16% of value in a market that's been riding the AI wave for two years straight. This isn't a minor correction. It's the market asking hard questions about who's actually going to pay for all this compute.
The proximate cause is boring: interest rates. But the deeper story is about how AI companies have financed their growth. They borrowed billions at near-zero rates to build GPU clusters and data centers. Now rates are up, refinancing looks expensive, and the math on these infrastructure plays is getting uncomfortable.
"Rising borrowing costs could hinder AI sector growth, increasing vulnerability to economic shifts and impacting future innovation."
CoreWeave is the poster child for this squeeze. The cloud compute provider that went public riding AI demand just watched its shares plunge 16% as bond yields approached 15%. When your debt costs 15%, you need extraordinary revenue growth just to tread water. And traders know it: they're buying puts on CoreWeave, betting the stock has further to fall.
Even with Deutsche Bank maintaining a buy rating, the message from bond markets is clear: high leverage in tech infrastructure is back to being risky. The free money era masked how capital-intensive this business really is. Building AI infrastructure means:
- Buying or leasing GPUs at $30K-$50K per chip
- Constructing or leasing data center space with massive power requirements
- Signing long-term power contracts before you have matching revenue contracts
- Hoping your customers' AI projects actually ship and scale
That last part is the real gamble. Debt-funded AI ventures assumed continuous, aggressive demand for compute. But if enterprise AI adoption slows, or if the big cloud providers (AWS, Azure, GCP) win most of the workloads, the neoclouds are stuck with expensive hardware and expensive debt.
The 8-day chip stock decline suggests investors are repricing the entire AI infrastructure stack. Nvidia, AMD, and the chip designers sold tens of billions in GPUs to companies that financed purchases with cheap debt. If those buyers stumble, future orders dry up. The semiconductor companies don't carry the debt, but they carry the revenue risk.
The Implication
Watch the neoclouds and second-tier AI infrastructure plays. If CoreWeave's debt costs stay at 15%, expect restructuring or fire sales. The companies with strong balance sheets and actual revenue from running AI workloads will be fine. The ones that bet everything on "if we build it, they will come" are about to find out what happens when the cost of capital matters again.
For anyone building in the agent economy: this is a reminder that infrastructure has to pencil out. AI agents need compute, but compute at any price isn't a business model. The shakeout in AI infrastructure stocks is part of the maturation process. Some of these companies will fail. The ones that survive will have proven unit economics, not just growth projections made when money was free.