Thirty billion dollars. That's more than most tech companies are worth, period. And Anthropic wants to raise it in a single round, at a valuation north of $900 billion.

Bloomberg reported this week that the AI safety company is in early talks to close the round by the end of May. For context, the entire Oracle market cap is around $450 billion. Anthropic is pitching more than twice that — for a company with no meaningful revenue, no dominant market share, and a models that cost more to run than anyone else's.

But that's the point. This isn't a normal market. The AI inference wars have turned compute into the new oil, and Anthropic is betting that safety differentiation will justify a premium so steep it makes Azure look cheap.

The secondary market crackdown

What's telling is what Anthropic did before announcing the raise: they named eight unauthorized secondary market sellers — Hiive, Forge Global, and others — and told investors any purchases there are void. They want to control who owns this company, and they want institutional money, not retail flippers. When you're raising at a $900B valuation, you can't have your stock being traded by people who don't understand what they're buying.

The enterprise push

The same week, Anthropic announced 12 Claude plugins for the legal sector, including a "commercial counsel" tool for vendor agreement review and a bar exam study tool. This is textbook enterprise play — lawyers bill by the hour and have an insatiable appetite for anything that makes review faster. It's also a sharp contrast to OpenAI, which has focused on consumers and developers. Anthropic is building the boring stuff that prints money.

The government contract

The US Department of Defense is deploying Mythos — Anthropic's model — to find and patch software vulnerabilities across the federal government. That's a massive vote of confidence from the one buyer that doesn't care about price.

What comes next

If this round closes at $900B+, Anthropic will have done something no AI company has ever done: raise more money in one go than the market thinks they're worth. It'll either validate the thesis that safety-first AI is the only kind of AI that matters, or it'll be the most expensive lesson in venture history.

Either way, the cloud providers just got a new competitor with more cash than they do.