Ramp's AI Index shows Anthropic widening its lead on OpenAI
Anthropic's share of US business AI spend grew from 41 to 43.5 percent between May and July. OpenAI moved from 39 to 39.7. The gaining-on-Anthropic headlines rest on a separate, unquantified claim about August.
Ramp's own numbers say Anthropic gained more ground than OpenAI. The headlines say the opposite.
Ramp's August 2026 AI Index tracks real transaction and token-spend data across more than 70,000 US business customers. TechCrunch reported on it August 20 under a headline about OpenAI gaining ground on Anthropic. Ramp's own May-to-July numbers say the opposite: Anthropic's lead grew, it did not shrink.
The hard numbers
Ramp's index runs on card transactions, bill pay, and token spend inside its own platform, not a survey. Its customer base skews toward tech and services companies, so treat it as a large real-spend sample, not a census of corporate America. The topline share of US business AI spend, also covered by VentureBeat and SaaStr, moved like this between May and July 2026:
- Anthropic: 41 percent in May, 43.5 percent in July. A gain of 2.5 percentage points.
- OpenAI: 39 percent in May, 39.7 percent in July. A gain of 0.7 percentage points.
- Anthropic's lead over OpenAI: 2.0 points in May, growing to 3.8 points in July.
Why headlines say OpenAI is gaining anyway
The gaining-on-Anthropic framing does not come from the numbers above. It comes from a separate comment: Ramp economist Ara Kharazian described OpenAI's growth as accelerating in Q3-to-date data, meaning partial August figures that were not published as a number. TechCrunch reported the characterization, not a percentage.
That is worth taking seriously. Kharazian has access to the underlying data and no reason to invent a trend. But it is a different kind of claim than the May-to-July numbers: unquantified, partial-quarter, and reported secondhand through a single comment. The two claims deserve different confidence levels, not the same headline.
Why a build studio cares
We build AI workflows and custom agents for clients, and we pick the model per project: something cheap and fast for a chat feature, something long-context for a document pipeline. The number in Ramp's data that matters to that decision is not who is ahead. It is that a vendor's share of spend moved by low single digits over three full months, in a market where switching providers costs an API key change, not a migration.
That is a low-switching-cost market, and it means vendor spend share is not a stand-in for lock-in the way it would be for a database migration. A 2 point lead becoming a 3.8 point lead in one quarter is real movement, but it is not evidence that this quarter's leader stays the leader. We build with that assumption already: pick on today's price and capability, keep the fallback provider tested, and do not treat any vendor's current position as fixed.
Next step: read Ramp's AI Index directly before trusting anyone's summary of it, this one included. TechCrunch's write-up has the gaining-ground framing this piece is checking against the index's own numbers. If you are choosing a model provider for a client workflow and want the switching-cost math done properly, write to hello@gattyworks.com.