What it means that OpenAI is closing Anthropic's lead among US companies — August 21, 2026
· CompaniesAutomation
Flash edition of the Radar: Ramp's index of 70,000 US companies gives Anthropic 43.5% and OpenAI 39.7% in July, but OpenAI is growing faster so far in Q3. What that says about the loyalty —and distribution— of your AI spend.
Flash edition. Anthropic remains in the lead among US companies paying for AI, but yesterday's data is about the derivative, not the snapshot: so far this quarter, OpenAI is growing faster. A supplement to this morning's Radar.
What happened
Ramp —expense management for over 70,000 US companies— published its monthly index: in July, 43.5% of those companies paid for Anthropic subscriptions or tokens, up 1.1 points in the month, compared to 39.7% for OpenAI, which rose just 0.23 (Source). TechCrunch added the nuance yesterday that changes the reading: according to Ara Kharazian, economist at Ramp, so far in the third quarter OpenAI is growing ahead of Anthropic; in May the picture was 41% vs 39% (Source). Their explanation is product-driven: GPT-5.6 Sol "is really good, increasingly the choice of developers," while Fable 5 "disappointed in adoption and real application" due to price and regulatory data retention requirements. Beneath this, a third movement: 6.1% of companies using AI are already buying via open-source or Chinese model platforms (Source).
Why it matters
The fact that the gap opens and closes in weeks makes whoever signs a one-year contract uncomfortable: enterprise AI spending is not sticky. Companies move when a better model comes out, not when renewal is due. And there is a figure that goes straight to your bill: in July, Fable 5 accounted for 6% of tokens purchased from Anthropic but 11.4% of the money; GPT-5.6 Sol, 25% of tokens and 23% of OpenAI spending. The expensive model takes a much larger portion of the budget than the work it actually performs.
For your company
Three concrete things. One: break down your bill by model, not by provider, and see what percentage the premium ones take and for which tasks; if you repeat the Fable 5 pattern —low volume, high cost— there is your savings without touching quality. Two: set up a bank of 20 or 30 real cases with the response you consider good; turn the next launch into a ten-minute decision instead of a blind migration and decouple the cost of your agents from the provider of the moment. Three: add data retention to your requirements sheet, alongside price: according to Ramp, it is already slowing down real adoption. If you prefer that layer of portability and measurement set up once and for all, an AI consultancy can have it ready in weeks.
Frequently Asked Questions
Who is winning, Anthropic or OpenAI?
Anthropic by level: 43.5% of US companies versus 39.7% for OpenAI in July. OpenAI by pace: it is growing faster this quarter. These are two different metrics and neither is a forecast.
Should I switch providers based on this data?
No. They measure adoption among US companies with corporate cards, not quality for your specific case. Switch if your own tests say so; other people's market share doesn't pay your bill.
What if my company uses neither of the two?
It is the growing third way: 6.1% of companies with AI are already using open-model platforms. It's cheaper at volume, in exchange for the operation becoming yours.