이미지: TechCrunch AI
Summary
- Ramp's tally of paid enterprise users in the US showed the ranking flip in May, with Anthropic at 41% and OpenAI at 39%, before the gap widened in July to roughly 44% for Anthropic and 40% for OpenAI
- Ramp economist Ara Kharazian said OpenAI has been growing faster than Anthropic so far in Q3, noting that GPT-5.6 Sol is increasingly becoming a go-to choice among developers
- Anthropic's top-tier model, Fable 5, has seen slower adoption due to its price combined with a 30-day data retention requirement
- 5월 점유율(램프)
- 앤스로픽 41% · 오픈AI 39%
- 7월 점유율(램프)
- 앤스로픽 약 44% · 오픈AI 약 40%
- 조사 대상
- 램프 이용 미국 기업 7만 곳 이상
- AI 유료 이용 기업 비중
- 3월 50% 돌파, 7월 약 56%
- Fable 5 데이터 보관 요구
- 이용자 데이터 30일 보관 의무화, 이용자 반발 발생
- 조사 한계
- 아메리칸 익스프레스 등 타 지출관리 도구 쓰는 대기업은 미포함
The ranking flipped in May, and the gap widened by July
Neither OpenAI nor Anthropic has gone public yet, so neither discloses revenue. That means anyone trying to gauge the two companies' actual business performance has to look to third-party data. One such source is figures released by Ramp, a corporate card and expense management service. According to Ramp's tally, OpenAI — once dominant in both the enterprise and consumer markets — was overtaken by Anthropic in US enterprise paid-user share for the first time this past May. Anthropic stood at 41% and OpenAI at 39% at the time, and OpenAI has not reclaimed the lead since. By July, the gap had widened further, with Anthropic at roughly 44% and OpenAI at roughly 40%.
These figures are drawn from spending data across more than 70,000 US companies that use Ramp's billing and corporate card services. There are limitations: Ramp's customer base skews toward Silicon Valley startups, and large enterprises that use other expense management tools such as American Express aren't captured in this data. Still, given that it reflects actual payment data from tens of thousands of companies, it's considered a solid signal for reading market trends.
The gap appears to be narrowing again in Q3
| Period | Anthropic | OpenAI |
|---|---|---|
| May 2026 | 41% | 39% |
| July 2026 | ~44% | ~40% |
Ramp economist Ara Kharazian said that a closer look at the most recent data shows OpenAI growing faster than Anthropic in this segment since July, into Q3. He added the caveat that with a month still left in the quarter, the trend could shift again. Ramp did not disclose actual dollar spending figures, only the share percentages.
In a post on X, Kharazian said, "GPT-5.6 Sol has gotten really good and is becoming an option developers are choosing." He added that Anthropic's top-tier model, Fable 5, has fallen short of expectations in both adoption and actual usage due to its price and a data retention requirement imposed by regulators.
Why Fable 5 has stalled
Fable 5 is Anthropic's flagship model line, designed for narrower use cases rather than as a general-purpose chatbot. It drew some backlash after Anthropic announced it would be expensive and would require mandatory 30-day retention of user data. The requirement was largely a response to regulatory pressure, but it ended up being cited as one factor slowing adoption.
The overall market keeps growing
Even as the two companies fight over market share, the proportion of Ramp customers spending money on AI tools overall keeps rising. It passed 50% this past March and climbed to roughly 56% by July. Regardless of who's winning the share battle, this suggests both OpenAI's and Anthropic's absolute revenue have room to grow together.
Editor's take
What this data really shows isn't an Anthropic victory — it's low customer loyalty among enterprise buyers. Companies that dropped OpenAI for Anthropic in May appear ready to pack up and switch back the moment a new model comes out. That's part of the backdrop for why, on August 14, both companies cut prices by roughly 20% around the same time, kicking off a price war. Holding on to enterprise wallets now requires more than model performance — pricing has to move too.
In practice, this pattern tends to repeat: engineering teams switch providers every few weeks based on API costs and latency, while executives only see the consequences of those decisions in a bill a quarter later. A model like Fable 5, priced roughly twice as high, gets stuck at the budget approval stage no matter how strong its performance is. Conversely, a lower-priced model like GPT-5.6 Sol can clear the adoption bar based on an individual developer's decision alone. For domestic companies, this suggests it's safer to review model contracts on a quarterly basis rather than locking in annual terms. In particular, clauses tied to regulatory issues — like data retention requirements — are worth having legal review before adoption; it's cheaper than the alternative.
The thing to watch over the coming weeks is the final numbers at the close of Q3. If the OpenAI rebound Kharazian described holds through the end of August, the share battle between the two companies could settle into a pattern that flips every quarter. And that volatility itself is likely to raise questions among investors about the stability of both companies' enterprise revenue.
Comments