Anthropic passes OpenAI. Skepticism sells better than messianism.

30 versus 25

30 billion dollars in ARR against 25. That's Anthropic's April 2026 figure, compared to OpenAI's on the same metric. For the first time since generative AI hit the mainstream, the commercial number one changes name.

To get the scale of the shift, look at the velocity. Anthropic went from 1 to 30 billion ARR in fifteen months. A tripling in four months, from 9 to 30. 80% of revenue comes from the enterprise. More than 1,000 customers spend over a million dollars a year with them, against 500 in February.

The shop that runs no ads just lapped the one running ads everywhere.

Who Anthropic actually is

For readers who missed the pilot episode, the useful recap is short. Anthropic was founded in early 2021 by Dario Amodei and his sister Daniela, alongside Tom Brown, Jack Clark, Jared Kaplan and several others. They all came from OpenAI, which they left in late 2020 over a clear disagreement: the house wanted to accelerate scaling before resolving safety. They wanted the opposite.

Five years later, their main product is called Claude. Their most cited paper is Constitutional AI: a method to embed principles during training, rather than filter at output. Their valuation hovers around 380 billion dollars.

What goes unseen from them is as telling as what is seen. No DevDay, no Super Bowl spot, no Sora, no GPT Store, no keynote where the CEO pulls out a holographic phone.

When Dario Amodei speaks publicly, it is in a long-form essay on their website, or at a Senate hearing. It is never a demo that gets the CEO sweating through his shirt.

Meanwhile in Oakland

The contrast is sharper still in May 2026, because OpenAI is in court. Elon Musk's lawsuit goes after the trajectory of OpenAI's mission: non-profit at founding, capped-profit in 2019, capped LLC in 2024. The complaint talks about a "culture of lies" around the shift.

Tuesday May 12, Sam Altman is on the stand. Musk's lawyer, Steven Molo, asks if he is fully trustworthy, if he always tells the truth. Altman hesitates, suggests a detour through his sincerity, Molo brings him back to the question. Altman ends up conceding that at some point in his life, it happened that he did not tell the truth.

That's a sworn admission, from the CEO of the most visible company of the decade, in a trial probing the reliability of his leadership (the substantive issue in that trial is the non-profit to for-profit pattern across AI labs, not the point here). In parallel, the enterprise market votes with its checkbook. And it votes Anthropic.

The CIO doesn't want a show

To get why the market is picking the quietest shop, you have to step into an enterprise buyer's head.

When a CIO buys an AI platform, they are not wondering whether it will make them miss a train. They are wondering what happens if their model hallucinates in front of a client, leaks a GDPR data point, or gets named in a class action. What they are buying is insurance against worst-case scenarios. Not a transformation promise.

Anthropic sells exactly that dossier. Constitutional AI sounds boring at a consumer conference. In a buying committee with general counsel in the room, it is what unlocks the signature. Interpretability research papers don't make the TechCrunch front page. They are part of the package the CISO accepts to submit to the risk committee.

OpenAI, in parallel, makes noise. That's what its leadership decided. Spectacle conferences, demos that turn virtuoso, 850 billion valuation projections, public debates on imminent AGI. Excellent for press. Unsettling for the risk manager who has to sign a three-year contract.

When discretion becomes an argument

The 2030 projected training cost adds a final touch to the picture. OpenAI: around 125 billion dollars per year. Anthropic: around 30. Four times cheaper to generate the currently higher revenue. At equivalent ARR, Anthropic's unit economics is radically healthier.

The B2B market puts these two things on the same scale. Reputational risk on one side, unit economics on the other. And it picks.

For the general audience, this signal carries further than AI. If the company that speaks the quietest takes the lead on the market that actually matters, mainstream AI marketing changes status. The DevDay, the X-buzz demo, the Super Bowl spot become signs of a consumer strategy that is no longer converting at the top of the market. Discretion becomes a buying argument. Sobriety becomes a proof point.

The signal

Anthropic is not declaring victory. Dario Amodei has not tweeted his ARR. No triumphant press release. The figure came out in an internal financial publication, picked up by analysts. The posture stays coherent even at the moment of overtaking.

Sam Altman, meanwhile, is defending his company at the bar of an Oakland courthouse. Two scenes, two postures, one ranking that just changed.

Skepticism sells better than messianism. Not a coincidence. A market signal.