Anthropic's annualized revenue run rate has surpassed $30 billion, up from approximately $9 billion at the end of 2025 — a threefold increase in roughly three months that has no precedent in the history of enterprise software [1]. The company that built the most powerful AI model and refused to release it is now, by at least one measure, generating more revenue than OpenAI.

The number arrived Monday in a statement timed to a new infrastructure deal with Google and Broadcom for 3.5 gigawatts of next-generation TPU capacity, coming online in 2027 [1]. On Friday, CoreWeave announced a separate multi-year agreement to supply Anthropic with cloud computing for its Claude models, sending CoreWeave's stock up 13 percent [2]. The deals are enormous. The revenue figure is more so. And the tension between Anthropic's two public identities — the company too responsible to sell its most dangerous model and the company printing money faster than any software firm in history — has never been sharper.

When Anthropic raised $30 billion at a $380 billion valuation in February, it reported that more than 500 business customers were each spending over $1 million annually [1]. That number now exceeds 1,000. It doubled in under two months [3]. Claude Code, the company's agentic coding tool, has reached $2.5 billion in run-rate revenue on its own, more than doubling since the start of 2026 [4]. The growth is not coming from consumer subscriptions. It is coming from enterprises integrating Claude into their workflows at a pace that suggests less adoption than dependence.

The trajectory is almost cartoonish in its acceleration. Anthropic's annualized revenue was approximately $1 billion in January 2025. It hit $3 billion by April, $5 billion by August, $9 billion by December, $14 billion by February 2026, and $30 billion by April [4]. Dario Amodei, the company's CEO, has noted repeatedly that he has consistently underestimated the growth rate of his own business. He has been wrong every time in the same direction.

The competitive implications are significant. OpenAI, which is projecting $20 billion in annualized revenue for 2026 but losing roughly $14 billion against it, now finds itself in the unfamiliar position of being outpaced on revenue by a company with a fraction of its consumer brand recognition [4]. Anthropic has no chatbot with 800 million weekly active users. What it has is the enterprise market, where the pitch is different, the contracts are larger, and the buyer cares less about brand and more about whether the model can do the work.

This is where the safety thesis becomes interesting — not as principle but as strategy. Anthropic's founding story is a departure from OpenAI over safety concerns. Its public posture has consistently emphasized responsible development. Last week's Project Glasswing announcement — keeping Claude Mythos Preview locked behind a consortium of twelve corporate partners for defensive cybersecurity use only — was the most dramatic expression yet of a company voluntarily withholding its most advanced product [5]. "Too dangerous to release" is not a marketing slogan. But it functions like one.

The enterprise customers spending seven figures annually are not paying for safety. They are paying for capability. But Anthropic's safety brand gives procurement committees something that OpenAI's "move fast" reputation does not: cover. An enterprise CIO who chooses Claude can point to Anthropic's published safety research, its Responsible Scaling Policy, and its conspicuous restraint with Mythos Preview as evidence that the vendor takes risk seriously. In a regulatory environment where AI governance is tightening in Europe and contested in Washington, that cover has dollar value [6].

The IPO compounds the thesis. Anthropic executives have discussed going public as soon as Q4 2026, with investment banks expecting a raise exceeding $60 billion — potentially the second-largest IPO in history behind SpaceX [7]. Bankers and lawyers largely expect Anthropic to list before OpenAI. The reasoning is instructive: public market investors prefer Anthropic's enterprise-heavy revenue mix, its lower customer concentration risk, and — crucially — its narrative. A company that says "we are building the most powerful AI in the world and we are being careful about it" is a more investable story than a company that says "we are building the most powerful AI in the world and we are shipping it to everyone as fast as possible." Both companies are doing approximately the same thing. The framing is what differs.

None of this means the safety commitment is insincere. Anthropic's research on constitutional AI, its interpretability work, and its willingness to absorb the costs of withholding Mythos Preview are substantive contributions to the field. The hundred million dollars in usage credits committed to Project Glasswing is not trivial for a company that burns compute at a prodigious rate [5]. But sincerity and strategy are not mutually exclusive. The same company can genuinely believe that frontier AI models pose catastrophic risks and also recognize that saying so publicly, loudly, and often is the most effective competitive positioning available in 2026.

The Pentagon's designation of Anthropic as a potential supply-chain risk — a dispute the company is fighting in federal court — has not slowed the growth [1]. If anything, it has reinforced the brand. An enterprise customer watching Anthropic challenge the U.S. government over AI safety guardrails sees a company with principles worth defending. The legal fight is expensive. It may also be the most effective marketing campaign Anthropic has ever run.

Thirty billion dollars in annualized revenue. A model too dangerous to sell. An IPO that could value the company at nearly half a trillion dollars. The pitch to investors and the pitch to regulators are the same words in different fonts. Whether that makes Anthropic the most principled company in technology or the most sophisticated is a question the market will answer in Q4.