Anthropic Leads OpenAI in Business AI Adoption—but OpenAI Is Closing Fast

Ramp’s latest business-spending data points to a close and changeable contest between OpenAI and Anthropic. Anthropic held the larger share of Ramp’s paying business customers in July 2026, but OpenAI was growing faster in the third quarter to date. The result is best read as an early adoption signal—not a definitive measure of the entire U.S. enterprise AI market. 367

The July leader was Anthropic

Anthropic first moved ahead of OpenAI among Ramp’s paying business customers in May. At that point, roughly 41% of Ramp businesses paid Anthropic, compared with about 39% paying OpenAI. 5

By July, Anthropic’s share had reached 43.5%, while OpenAI’s stood at 39.7%, leaving Anthropic with a lead of 3.8 percentage points. 31314

Those figures measure the percentage of Ramp businesses paying each provider. They are not mutually exclusive shares of revenue: a company can pay both Anthropic and OpenAI. That distinction matters because enterprise buyers may use different models for coding, research, customer service, or internal workflows rather than choosing one provider exclusively.

OpenAI’s momentum is the more important change

Although Anthropic remained ahead in the latest full-month figures, Ramp said OpenAI was adding paying business users faster than Anthropic in Q3 to date. That suggests the July gap may not be a stable ranking. 67

The data therefore tells two stories at once:

• Anthropic won the latest adoption snapshot, with 43.5% of Ramp businesses paying for its products versus 39.7% for OpenAI. 3
• OpenAI had stronger recent growth, indicating that it was beginning to narrow the distance after Anthropic’s spring advance. 67

No source provided here gives a precise Q3 growth rate for either company, so the evidence supports a direction of travel rather than a forecast about who will lead next.

Model economics and privacy may be influencing buyer choices

Ramp’s model-level data offers clues about why OpenAI may be regaining momentum, though it does not establish causation.

In July, OpenAI’s GPT-5.6 Sol accounted for 25% of tokens purchased from OpenAI and 23% of OpenAI model spending. Anthropic’s newer Fable 5 represented only 6% of Anthropic tokens and 11.4% of Anthropic model spending. Fable 5 generated roughly 75% as much attributed spending as Sol. 317

One possible explanation is that developers weigh price and usability alongside benchmark performance. Ramp’s comparison found Fable 5 to be substantially more expensive than Sol, despite Fable’s stronger reported performance in some evaluations. 3

Data handling may also matter for some enterprise buyers. Reporting said Anthropic’s most advanced models required business customers to retain data for 30 days, while the company planned a later system that would give customers more control by allowing the retention infrastructure to run on their own cloud. 3233

That policy could be a disadvantage for regulated or privacy-sensitive organizations, but it should not be treated as a proven explanation for Ramp’s adoption figures. Transaction data can show what companies paid for; it cannot show whether price, performance, privacy, procurement, or another factor caused the decision.

What Ramp’s sample can—and cannot—show

Ramp’s analysis uses aggregated, anonymized card and bill-pay transactions from more than 70,000 U.S. businesses. It is a substantial spending dataset, but it is not a census of American companies or a complete record of enterprise AI contracts. 318

Several limitations shape how the figures should be interpreted:

• Spending routed through cloud marketplaces, resellers, bundled software, separate procurement systems, or invoices outside Ramp may not appear in the data.
• Ramp’s customers—and especially companies buying model tokens directly—may be more technology-oriented than the typical U.S. business.
• The token-spending sample used for the Fable 5 and GPT-5.6 Sol comparison is more technology-focused than Ramp’s broader AI adoption sample. Ramp has cautioned that actual Fable adoption may therefore be lower than the estimate in that comparison. 3
• The percentages describe adoption among Ramp customers, not national enterprise market share. 43

There is also a reporting difference across Ramp updates: earlier July figures cited by Ramp-related reporting put Anthropic at 42.4% and OpenAI at 39.5%, while later coverage reported 43.5% and 39.7%. 1113 The broader conclusion is consistent across the figures—Anthropic led in July, and OpenAI was growing faster in Q3—but the exact percentage depends on the update and methodology used.

Enterprise AI spending is growing, not simply changing hands

The rivalry is unfolding within a larger expansion of business AI adoption. Paid AI use among Ramp businesses crossed 50% in March, reaching 50.4%. 22

By July, nearly 56% of Ramp businesses were paying for AI services. 36 That means the OpenAI–Anthropic contest is not merely a fight over a fixed pool of customers. More companies are entering the market while existing buyers experiment with multiple providers.

This growth also helps explain why adoption percentages and spending can tell different stories. A provider may add customers without capturing the largest share of usage, while a smaller group of heavy users can drive a disproportionate amount of model spending.

The bigger takeaway: enterprise AI is still negotiable

The sequence from Anthropic’s rise in May to OpenAI’s faster Q3 growth suggests limited vendor lock-in at this stage. Businesses appear willing to test, combine, or reconsider providers as model quality, pricing, privacy terms, and developer tools change. 67

That does not mean corporate AI spending is weakening. It means the market is expanding while vendor loyalty remains unsettled. For buyers, the practical implication is that choosing a model may be an ongoing portfolio decision rather than a permanent commitment to one lab.

For OpenAI and Anthropic, the July figures offer no final verdict. Anthropic still had the larger footprint in Ramp’s sample, but OpenAI’s recent acceleration shows how quickly a new model, a pricing difference, or a change in data policy can reshape the competitive picture.