The money is getting nervous, and that's when things get interesting.

The Summary

  • Barry Ritholtz tells Bloomberg AI is transformational but picking winners is getting harder — CoreWeave's selloff is exhibit A
  • Investor confidence in AI valuations is cracking as financing concerns mount
  • The real test: who's building businesses versus who's building hype on borrowed capital

The Signal

CoreWeave, the GPU infrastructure company that was supposed to be the picks-and-shovels play for the AI gold rush, is selling off. That matters because CoreWeave isn't some chatbot startup burning through venture money. It's the backbone company that rents compute to everyone else. When the infrastructure layer wobbles, the whole stack gets scrutinized.

Ritholtz's read is straightforward: AI is real, the transformation is happening, but the market got ahead of itself on who captures value. The current selloff isn't about AI being fake. It's about investors realizing they don't actually know which companies will still be standing when the music stops.

"AI remains transformational, but picking the winners is the hard part."

The financing angle is where this gets sharp. A lot of AI infrastructure got built on cheap capital and the assumption that demand would justify any price. CoreWeave raised billions. Nvidia became a two-trillion-dollar company. Everyone assumed the compute requirements for training and inference would create infinite demand for GPUs and the companies that warehouse them.

But here's what's shifting:

  • Model efficiency is improving faster than expected, meaning less compute per unit of value
  • Open-source models are closing the gap with proprietary ones, commoditizing performance
  • Hyperscalers like Microsoft and Google are building their own chips, cutting out the middlemen

That creates a margin squeeze. If you're CoreWeave, you need utilization rates to stay high and prices to stay premium. If customers can get similar performance from cheaper alternatives or need less compute overall, your billion-dollar data centers start looking like expensive real estate in a declining market.

The Implication

Watch who's building defensible moats versus who's riding the wave. The companies that survive this aren't the ones with the most GPUs or the flashiest demos. They're the ones solving real problems that justify the compute cost, or the ones so far down the cost curve they can survive commoditization. If you're betting on AI infrastructure, ask whether the business model still works when compute gets 10x cheaper. If you're building with AI, now's the time to prove you can deliver value that justifies the burn rate. The easy money phase is over.

Sources

Bloomberg Tech