OpenAI Claws Back Ground on Anthropic in Business AI, Corporate Card Data Shows

Anthropic still leads on how many US companies pay for its models, but OpenAI's newest release is winning the fight over usage and dollars

SAN FRANCISCO, August 21: OpenAI has started narrowing the gap with Anthropic among American business buyers, according to spending data from corporate card and expense platform Ramp, reversing part of a slide that cost the ChatGPT maker its lead earlier this year.

Neither company publishes financial statements, and both have filed confidentially for public listings. That leaves payment data as one of the few outside windows into who is actually winning corporate wallets.

The Share Gap Is Narrowing, Not Closing

Anthropic remains ahead. <cite index="48-1">In July, 43.5% of US businesses paid for subscriptions or tokens from Anthropic, up 1.1 percentage points month over month, while OpenAI rose only 0.23 percentage points to 39.7%, underperforming overall AI adoption. xAI posted its fastest growth since July 2025, adding 0.94 percentage points to reach 4% of businesses.</cite>

The crossover happened in the spring. <cite index="51-1">Anthropic's adoption rose 3.8% in April to 34.4% of businesses while OpenAI's fell 2.9% to 32.3%, the first time more American businesses paid for Claude than for ChatGPT. In April 2025, OpenAI had commanded roughly 32% of business AI adoption against under 8% for Anthropic.</cite>

<cite index="50-1">By July, Anthropic's strength was concentrated in high-adoption verticals such as finance and technology.</cite> TechCrunch reported that the Ramp dataset now covers more than 70,000 American businesses spending billions through its bill pay and card products, and that Anthropic first took the lead among these customers in May at 41% against OpenAI's 39%.

Ramp economist Ara Kharazian said OpenAI is growing faster than Anthropic among this segment in the third quarter so far. <cite index="50-1">One tally put OpenAI's quarter-over-quarter enterprise growth rate at 82 against Anthropic's 76 as of mid-August.</cite> A month of the quarter remains, and Ramp declined to release dollar figures, sharing percentages only.

Adoption and Intensity Are Telling Different Stories

Counting logos is one measure. How deeply those companies use a model is another, and the second one favours OpenAI right now.

<cite index="50-1">OpenAI's GPT-5.6 Sol accounted for 25% of tokens and 23% of corporate spend in July, while Anthropic's Fable 5 represented 6% of tokens and 11.4% of spend.</cite> More companies are signing up with Anthropic, but OpenAI's newest release is absorbing a disproportionate share of actual workload.

Kharazian attributed the shift partly to model quality, posting on X that GPT-5.6 Sol has increasingly become the developer default, while Fable 5 disappointed on adoption because of its price and data retention terms.

The Retention Fight Is Now Open Warfare

That second point deserves precision, because the retention policy was not imposed on Anthropic by regulators.

<cite index="63-1">Anthropic introduced the policy itself, effective June 9, 2026, for what it calls covered models. The company's stated reasoning is that some attacks are visible only across multiple requests: best-of-N jailbreaking sends hundreds of slight prompt variations hoping one succeeds, and campaigns such as state-sponsored espionage surface only when classifiers can look across many requests at once.</cite>

<cite index="57-1">Under the terms, prompts submitted to and outputs generated by covered models are retained for 30 days on every platform where those models are offered, including for commercial customers who had zero data retention agreements in place.</cite> <cite index="62-1">Anthropic said at the time that the retained data would not be used to train its models.</cite>

Enterprise reaction was fast. <cite index="60-1">Microsoft restricted internal Fable 5 use pending legal review, and in GitHub Copilot the model was disabled by default, requiring Enterprise and Business administrators to opt in explicitly.</cite>

OpenAI moved on the opening this week. <cite index="59-1">The company began testing a system called Private Safety Processing with early customers, designed to identify misuse patterns across related interactions while preserving zero data retention. OpenAI says the system sends back a narrowly defined safety signal without exposing the underlying prompts or responses, with customer data able to remain on customer-controlled infrastructure or be stored under encryption keys the customer holds. A broader rollout and a technical white paper are planned for September.</cite>

<cite index="62-1">Bloomberg reported on August 20 that Anthropic is planning changes to the retention policy for its advanced models, citing a person familiar with the matter who said the system had been in development for months. Anthropic declined to comment.</cite>

The privacy contrast is less clean than it looks from either side. <cite index="60-1">On January 5, 2026, US District Judge Sidney H. Stein affirmed an order requiring OpenAI to produce 20 million de-identified ChatGPT logs in consolidated copyright litigation, rejecting the company's privacy objection on the reasoning that users had voluntarily submitted their communications. A contractual retention promise does not put stored data beyond the reach of a court.</cite>

Revenue Runs the Other Way

Market share among Ramp's customers and top-line revenue are pointing in opposite directions.

<cite index="68-1">Anthropic told investors that its annualized revenue run rate hit $65 billion at the end of July, roughly a sevenfold increase year on year, according to CNBC, which confirmed a figure first reported by Bloomberg. In May the company put its run rate above $47 billion, against roughly $10 billion in revenue for all of 2025. OpenAI's annualized run rate recently reached $40 billion.</cite>

<cite index="69-1">Investors expect Anthropic to finish 2026 somewhere between $100 billion and $120 billion, the Financial Times reported. OpenAI has doubled its revenue to $40 billion from $20 billion at the end of 2025, according to Bloomberg. The two companies may calculate the metric differently.</cite>

<cite index="68-1">Anthropic filed its prospectus confidentially with the Securities and Exchange Commission in June and has held preliminary meetings with potential investors, working against a $965 billion valuation.</cite>

The Pie Is Still Expanding

Both companies can grow revenue while trading share, because more companies keep buying. TechCrunch reported that the share of Ramp customers paying for AI passed 50% in March and reached nearly 56% by July.

Cheaper alternatives have not displaced either lab. <cite index="53-1">In June, 5.8% of AI-spending businesses used model-serving platforms, Ramp's proxy for open-source and Chinese model adoption, up from 4.5% in January. Those firms spent a median of $248 per employee on AI that month against $10.59 for the median AI-spending business. Among them, 85.8% also used OpenAI, 93.2% used Anthropic, and 96.4% used at least one of the two.</cite> <cite index="48-1">By July that figure had edged up to 6.1%.</cite>

The Ramp panel has known limits. It skews toward technology companies, and it excludes large enterprises that run spend management through providers such as American Express. What it captures is the segment most willing to switch, and the past four months suggest that segment will move on the strength of a single model release.