While everyone was watching OpenAI, Anthropic quietly became the company holding up the entire AI trade — just as the foundation started cracking.

The Summary

  • Anthropic launches tomorrow with a $90B IPO — the largest ever — while AI infrastructure spending faces unprecedented scrutiny and Nvidia's CEO resigned over a spy scandal
  • Demand hit 7x available shares despite the sector teetering: data center backlash in midterms, circular spending questions, junk bond financing for infrastructure
  • The safety-focused OpenAI alternative spent years in the shadow, then Claude Opus 4.5 changed everything last November

The Signal

The timing is almost comically terrible, which means it's either brilliant or catastrophic. Anthropic is going public at the exact moment when AI infrastructure economics are getting their first real stress test. Jensen Huang's exit from Nvidia over an alleged Chinese spy affair isn't just tabloid fodder — it's the removal of the chief evangelist and capital allocator for the entire AI hardware stack. And Nvidia was already considering pulling back spending commitments before the scandal broke.

The midterm data center referendums tell you something important about where public sentiment is heading. When voters in multiple states explicitly vote to slow data center development, you're watching NIMBYism meet energy grid reality meet AI hype fatigue. The power demands are real. The local infrastructure impacts are real. And the "trust us, this will create jobs" pitch is wearing thin when most of these facilities employ dozens of people to serve billions in cloud compute.

"If this is a bubble, about now would be the time for it to pop."

But Anthropic's 7x oversubscription says the smart money still believes. Here's why that's both impressive and concerning:

  • The company spent years as the "safety-focused" alternative, which in Silicon Valley terms meant "slower and more cautious"
  • Claude Opus 4.5's November release apparently changed the competitive landscape enough that institutional investors are willing to bet $90B
  • This IPO isn't just about Anthropic — it's about keeping the entire AI infrastructure trade alive through what could be a rough 2027

The circular spending problem is real. AI companies buy compute from cloud providers who buy chips from Nvidia who invests in AI companies who buy more compute. When one link weakens, the whole chain rattles. Major data center players loading up on junk bonds to finance expansion is a classic late-cycle signal. You see this pattern in every infrastructure boom: easy money, massive buildout, leverage piling up, then someone asks "wait, what's the actual revenue model here?"

The Implication

If Anthropic's IPO succeeds, it buys the AI sector another 12-18 months to prove the economics work. If it stumbles, you're looking at a repricing across the entire stack — from Nvidia down to every startup that raised at a "AI-company premium" valuation. Watch the first week of trading. Watch whether institutional investors hold or flip their positions. And watch whether Anthropic can articulate a path to profitability that doesn't depend on infinite compute scaling.

For anyone building in this space: the era of "build cool AI stuff and figure out the business model later" is ending. The capital is still there, but it's getting more expensive and more skeptical. The winners in the next phase will be companies that can show actual unit economics, not just impressive demos and "we're pre-revenue but growing fast" narratives.

Sources

The Atlantic Tech