Anthropic Passed OpenAI in Business Adoption — Now Read the Second Chart

For the first time, more US businesses are paying for Anthropic than for OpenAI. That is the headline from Ramp’s AI Index, which tracks anonymized corporate-card and bill-pay spending across the companies on its platform: Anthropic reached 34.4% business adoption, edging past OpenAI’s 32.3%. A year earlier, OpenAI sat near 32% and Anthropic was under 8%. So in twelve months Anthropic roughly quadrupled its business footprint while OpenAI’s barely moved.

It is a real milestone, and TechCrunch, Axios, and The Decoder all covered it as one. But if you are the person who actually has to choose a model vendor — not write about the race, but live with the decision for the next two years — the leaderboard flip is the least useful part of the story. The useful part is what the number is measuring, what a second dataset says, and why both can be right at once.

What the Ramp number actually counts#

Ramp’s index is a spending signal. It answers one question well: of the companies on Ramp’s platform, what share have any paid relationship with a given AI vendor in a given month. That makes it an excellent early indicator of breadth — how widely a vendor has gotten its foot in the door — and a leading one, because card spend shows up before formal procurement does.

What it does not measure is how much those companies depend on the vendor. A 20-seat Claude Code pilot on one team’s card and a company-wide rollout register the same way: one company, adopted. So “34.4% adoption” means roughly one in three businesses is paying Anthropic something — not that one in three runs its core workflows on Claude.

The growth engine matters here too. Ramp and its commentators are unusually specific: nearly all of Anthropic’s gain traces to a single product, Claude Code. That is not a knock — Claude Code is a genuine category leader and the reason a lot of engineering orgs put Anthropic on a card at all. But it does shape what the win means. Anthropic’s business surge is, to a first approximation, a coding-tool surge. A vendor whose footprint rests heavily on one product has a different risk profile than one spread evenly across many.

Now read the second chart#

Here is where a single dataset misleads. A separate IDC survey from March 2026 asked a different question — not “do you pay them” but “do you use them extensively” — and got a different answer. Only about 19% of organizations reported extensive Claude use, trailing both OpenAI and Google by a wide margin on that measure.

Those two findings are not contradictory. They are measuring different things:

• Ramp measures breadth of adoption — who has started paying. Anthropic leads.
• IDC measures depth of use — who has gone all-in. OpenAI and Google lead.

Stack them and the real picture appears: Anthropic is winning the first purchase faster than anyone, largely on the strength of Claude Code, while incumbents still own the deepest deployments. Whether Anthropic’s breadth converts into depth — pilots becoming platforms — is the question neither chart answers yet. It is the only question that matters for the 2027 version of this story.

The revenue figure has the same trap#

The pattern repeats one level up, in the numbers used to value these companies. Anthropic has reported run-rate revenue above $30 billion with more than 1,000 customers spending $1M+ a year — a steep, real climb. But the headline figure hides an accounting choice. Anthropic books cloud-partner payments as gross revenue; OpenAI reports net of its Microsoft arrangement. Bank of America estimated Anthropic’s cloud payments at roughly $6.4 billion in 2026 — the size of the gross-versus-net gap.

That does not make the revenue fake. It means a side-by-side “Anthropic vs OpenAI revenue” comparison is not apples-to-apples until both report on the same basis, which an eventual IPO and SEC harmonization would force. The reflex worth building is the same one the adoption charts demand: before you compare two numbers, check they are counting the same thing.

What this changes for a vendor decision#

If you are picking — or re-picking — a primary model vendor, the takeaway is not “switch to whoever topped this month’s index.” It is to stop treating any single ranking as the decision and to run your own version of the second chart.

• Separate breadth from depth in your own org. “We have Claude on a card” and “our production workflows depend on Claude” are different commitments with different switching costs. Know which one you are actually making before you call it a standard.
• Score on your tasks, not the leaderboard. Ramp’s index is a market signal, not a capability verdict. The only ranking that should move your decision is your own eval harness on your own workloads, re-run when models change.
• Watch the concentration. Anthropic’s lead leans heavily on Claude Code. If your dependency leans on the same single product, your fortunes are correlated with one product line’s roadmap and pricing — worth knowing before it is a surprise.
• Keep an exit. VentureBeat’s own framing of the milestone flagged the threat underneath it: some of the fastest-growing vendors on Ramp are cheap open-model inference platforms. If cost becomes the dominant purchase criterion, both frontier labs face pressure from the commodity layer below them. An architecture that can route to a cheaper model when the economics shift is worth more than brand loyalty to whoever is winning today.

The takeaway#

Anthropic overtaking OpenAI in business adoption is a genuine inflection — twelve months ago it was not close. But the headline is a breadth metric riding largely on one product, and the depth metric still points the other way. Read both charts, not one. For a vendor decision, the lesson is method, not allegiance: define whether you are buying breadth or depth, benchmark on your own tasks rather than someone’s spending index, and keep a path to switch when the numbers underneath the headline move — because in this market, they move every quarter.