Anthropic just overtook OpenAI in business spending for the first time, and its latest fight with the Trump administration may end up helping rather than hurting it. That’s the read from TechCrunch AI, citing fresh data from corporate spending platform Ramp. The short version: getting branded too dangerous to use is turning into a sales pitch.
Here’s what happened, in order.
A month that’s hard to top
Anthropic closed May by passing OpenAI in market share of business AI spending. According to TechCrunch AI, the company also raised $65 billion at a $965 billion valuation, beating OpenAI on that front too, then opened June by filing confidential IPO paperwork on the back of its first profitable quarter.
Then Friday brought the twist. The Trump administration sent a letter demanding Anthropic block non-Americans, including its own employees, from its most advanced models: the limited-release Mythos 5 and a public version called Fable 5 that had shipped just three days earlier. The order effectively forced Anthropic to pull both from the market.
The White House cited an obscure export control directive, but TechCrunch AI reports the exact trigger is still unclear. The talk in the industry is that hackers slipped past Fable 5’s guardrails, which were built to keep the public from reaching Mythos-level capabilities. That model is reportedly so good at spotting security flaws in code that Anthropic marketed it as dangerous and held back its release.
Why a government fight isn’t slowing sales
This isn’t the first clash. Anthropic refused to let the government use its models for mass surveillance of Americans or fully autonomous weapons. In March, the Trump administration responded by labeling the company a supply-chain risk.
Business customers didn’t flinch. They leaned in. Ramp’s lead economist Ara Kharazian, who compiled the data, put it plainly to TechCrunch AI: “If anything, it’ll probably boost them. Anthropic’s best month on record, as far as business adoption, was the month that the Department of Defense labeled them a supply-chain risk. There’s a lot of aura that comes with your model specifically being named too dangerous to use.”
What stands out here is how the controversy doubles as validation. A model the government wants restricted reads, to a lot of buyers, as a model worth paying for.
The numbers behind the story
Ramp pulls from more than 70,000 businesses on its platform. A few figures from the report:
- Anthropic’s share of AI subscriptions paid by businesses rose 2.5 points in May to 41%.
- OpenAI sat at 39.5%, basically flat from the month before.
- OpenAI still leads heavily in consumer usage, per separate Sensor Tower data.
Most corporate spend goes to API calls, not subscriptions, covering token use for things like coding. Anthropic’s Claude Code has built a strong reputation as a coding tool. When Ramp can see model details, which is about a third of transactions, businesses are mostly spending on versions of Claude Opus, especially the newer ones. Opus is the model that came before Mythos and is still openly available, with Opus 4.8 landing in late May.
That timing matters. Mythos only reached limited users in April, and Fable 5 was live for a few days. So the models driving Anthropic’s record month are the ones still on the market, not the ones the government just forced offline.
What to watch next
The open question is the IPO. Public-market investors tend to get nervous about companies tangled in government fights, so the White House drama could complicate Anthropic’s plans to go public, as TechCrunch AI notes.
For practitioners, the takeaway is more practical. The tools you’re likely already using, the Opus and Claude Code lineup, are untouched by the ban. The pulled models were the bleeding edge, not the workhorses. If you’re building on Anthropic’s API, your stack keeps running.
The bigger signal is about how the market now reads risk. A “too dangerous” label used to be a warning. For business buyers, it’s starting to look like a feature. Whether that holds through an IPO filing is the next chapter worth watching.
Full details are at the original TechCrunch AI report.