OpenAI just raised $122 billion at an $852 billion valuation, and even the VCs cutting the checks are saying valuations are "punchy."

The Summary

  • OpenAI closed its largest fundraising ever at $122 billion, pushing its valuation to $852 billion while Anthropic hit $380 billion
  • Salesforce Ventures' managing partner openly called current AI valuations "punchy" on the record
  • The gap between enthusiasm and skepticism among those writing the checks is the real story

The Signal

When venture capitalists start using words like "punchy" to describe valuations they're actively participating in, pay attention. Paul Drews from Salesforce Ventures just said the quiet part out loud on Bloomberg: AI valuations are stretched. This matters because Salesforce Ventures is almost certainly in these deals or looking at similar bets. They're not sitting this out, but they're also not pretending the math makes traditional sense.

OpenAI at $852 billion is larger than most S&P 500 companies. For context, that's bigger than Costco, Abbott Labs, or Goldman Sachs. Anthropic at $380 billion is worth more than Netflix or Salesforce itself. These are foundation model companies with massive compute costs, uncertain revenue models, and existential questions about moats. The $122 billion raise isn't just big, it's a statement that the AI infrastructure layer is now being priced like winner-take-all public utilities before they've proven they can be profitable utilities at all.

The composition of this round tells you where we are in the cycle: tech giants hedging their platform risk, VCs chasing the category-defining outcomes, and now retail investors getting access to pre-IPO AI at nosebleed valuations. That last part is the canary. When retail gets invited to the private AI party, it usually means institutional players are looking to distribute risk, not concentrate conviction.

What Drews is really saying is: we know these numbers are wild, we're doing it anyway, and we're managing our exposure accordingly. That's not bearish on AI, it's realistic about what happens when an entire asset class gets repriced in 24 months.

The Implication

If you're building in the agent economy, this is your moment to define differentiation that isn't "we use Claude" or "powered by GPT-5." The foundation model layer is now a trillion-dollar oligopoly with unclear paths to return. The value is moving up the stack to companies that can prove unit economics with AI, not just deploy it. Watch for the smart money shifting from model makers to model users who can show margin expansion, not just automation theater.

For everyone else: when VCs call their own bets punchy, the risk-reward is changing fast. The AI boom is real. The AI bubble might be too.

Source: Bloomberg Tech