OpenAI narrows Anthropic's lead: build for the swap
Ramp data across 70,000+ US businesses shows the AI vendor lead flipping twice in four months. Design your stack so switching models is a config change.
New spend data says the AI vendor lead is not settling down. Anthropic passed OpenAI among US business buyers in May, extended that lead through July, and is now growing slower than OpenAI in the third quarter. Two flips in four months. If you are choosing a model provider this week, the useful takeaway is not who is winning. It is that model portability is the only position that survives the next flip.
What actually happened
TechCrunch reports on data from Ramp covering more than 70,000 American businesses that run spend through its corporate card and bill pay products.
The shape of it:
- May: Anthropic 41%, OpenAI 39% — the month Anthropic took the lead.
- July: Anthropic near 44%, OpenAI near 40%.
- Q3 to date: OpenAI is growing faster than Anthropic among these buyers, per Ramp economist Ara Kharazian, who credits the reception of OpenAI's newest model among developers.
The market itself is still expanding underneath the fight. More than 56% of Ramp's customers were paying for AI services by July, up from 50% in March. Both labs can grow while trading share.
Two caveats worth stating plainly, because they are in the reporting: the sample skews toward tech-industry companies, and it excludes large enterprises that run spend through American Express tooling. This is a real signal about mid-market software buyers, not a census of the whole economy.
Why vendor share is the wrong thing to optimize for
Businesses are flipping providers on model releases. That is the actual finding, and it is a compliment to buyers — they are switching because switching is now cheap enough to be worth doing when a better model lands.
It is only cheap if you built for it. We see two versions of the same client system:
The pinned build. One provider's SDK imported everywhere, provider-specific tool-calling formats, prompts tuned to one model's quirks, evaluation done by vibes. Switching means a rewrite, so the team stays on a model that lost its edge two releases ago and pays list price for the privilege.
The portable build. An OpenAI-shaped API surface behind one internal client, prompts in version control, and a small eval suite of your own real tasks. Switching is a base URL, a model string, and an afternoon of running the evals. The vendor's roadmap becomes their problem instead of yours.
The second one is not more expensive to build. It is the same work, done in a different order. And it means the next time the share numbers flip — which, on this evidence, is roughly a quarter away — you get to act on it instead of reading about it.
Key takeaways
- Ramp data across 70,000+ US businesses: Anthropic 41% vs OpenAI 39% in May, near 44% vs near 40% in July
- OpenAI is growing faster than Anthropic in Q3 to date, per Ramp's economist — the lead has now moved twice in four months
- AI adoption is still expanding: 56% of Ramp customers paid for AI by July, up from 50% in March
- Caveats: tech-skewed sample, and it excludes large enterprises using American Express tooling
- Build behind one internal client with your own eval suite so a provider switch is a config change, not a rewrite
Could you switch model providers this month if you wanted to? We build AI systems with a portable interface and your own evals, so a better model is an upgrade instead of a migration project. See how we architect AI systems, or tell us what you're locked into.
Sources: TechCrunch, Ramp AI Index.
- #ai-vendors
- #model-portability
- #openai
- #anthropic
- #ai-spend
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Rillet's $100M: put an audit trail under your agents
An AI-native ERP hit a $1B valuation by running agents inside the general ledger with human approval and full logging. That architecture is the product.
Read itRamp Router: your expense vendor now sells you inference
Ramp launched Router.com, a single API that routes every LLM call to the cheapest model that clears your bar. Free through 2026. Read the retention default first.
Read it