The AI lab behind the Claude chatbot reached 41 percent of business AI subscription spending on Ramp's corporate card index of 70,000 plus U.S. companies in May, the same month the Trump administration forced two of its models off the market.
In May, Anthropic, the AI lab behind the Claude chatbot, reached 41 percent of business spending on AI subscriptions across more than 70,000 U.S. companies on Ramp's index. That is a record high, up 2.5 percentage points from April, and ahead of OpenAI's 39.5 percent, which was flat, according to Ramp's AI Index. It is also the second consecutive month Anthropic has set the pace while the federal government has treated it as a security threat.
The chronology is what makes the data more than a coincidence. Anthropic's strongest run of enterprise adoption began in March 2026, the same month the Department of Defense added the company to a supply-chain risk list normally reserved for Chinese surveillance vendors and Russian antivirus firms. Two months later, the company raised $65 billion at a reportedly $965 billion valuation, filed confidential paperwork for an initial public offering, and posted its first profitable quarter. By June 12, a Trump administration letter invoking an export-control mechanism forced Anthropic to pull both Mythos 5, its most advanced model that had been in limited release since April, and Fable 5, a public-facing Mythos variant that had been on the market only days. The flagship is now off the table entirely.
None of that has dented the demand curve. As Ramp economist Ara Kharazian told TechCrunch, "There's a lot of aura that comes with your model specifically being named too dangerous to use." The color is the analyst's; the underlying mechanism is mechanical. Anthropic's Claude Code tooling and direct API calls, not consumer subscriptions, now dominate the spend mix on Ramp. The transactions are weighted toward later Claude Opus variants, per Kharazian's observation, and the buyers are wiring the model into production workflows rather than signing up for chatbot access.
That is the spine of the story. The customers Anthropic is winning in May are not the audience the Trump administration is signaling to. Government risk designations, export-control letters, and forced model withdrawals function as a kind of provenance for the frontier-AI buyer: a vendor the Pentagon is willing to fight has cleared a credibility bar that a friendlier lab has not. The friction is a feature, not a bug, for procurement teams who need a defensible reason to standardize on one model family for the next three years.
The honest caveat: one month of one dataset, corporate card spend on a single B2B platform, is correlation, not causation. Ramp's index measures U.S. business adoption. It does not capture global enterprise contracts, defense procurement, or the distribution advantage OpenAI still holds on the consumer side. ChatGPT leads Claude on consumer downloads and time spent per Sensor Tower's State of AI 2026 report, cited via TechCrunch. The June 12 letter landed late in the May measurement window, and the model withdrawals came after. The next two monthly prints will tell us whether the timing is the story or a coincidence dressed up as one.
If the pattern holds, the next round of frontier-AI disputes will look different. The commercial logic now points one direction: the more a government treats a model as too dangerous to sell, the more enterprise buyers treat that label as a procurement signal. Anthropic has stumbled into the most powerful go-to-market motion in artificial intelligence, a regulator who cannot stop talking about how unsafe the product is.