AI spend per employee slumped at top firms in August — summer doldrums or a warning sign?

The adoption of AI tools by businesses slowed in August, according to spending data at 70,000 companies collected by the payments company Ramp. The latest survey shows 56% of Ramp customers paid for AI products in August, rising just 0.4% from the month before.

This isn’t the first time Ramp’s metrics have shown adoption slowing down. Last year, the company’s AI index showed little to no growth in adoption between August and October, only to have growth pick up again as the year finished.

Still, the extreme pace of the AI buildout means even small slowdowns can be cause for concern. The gobsmacking investment in AI infrastructure by frontier labs and hyperscalers rests on the hope that there is plenty of revenue out there to pay it back. Thus far, usage has grown steeply, particularly as software engineers adopted agentic coding tools — but if that adoption slows down, revenue is likely to slow as well.

Ramp’s figures may overstate overall adoption, thanks to the company’s techy clientele: An ongoing U.S. Census Bureau survey of AI adoption updated on August 23 shows just 22% of businesses report using AI. Ramp’s survey isn’t necessarily representative of the market, but it’s one of the few direct spending datasets available and potentially a leading indicator.

To be sure, this data is from August, when much of the industry is on vacation. That may explain the doldrums. But there are other warning signs for companies that depend on token spend, per Ramp economist Ara Kharazian.

First, a major decline in AI spend per employee in the top 1% of AI-using firms in his sample, falling nearly 10% to $7,205. That may be the vacation-token factor, but it also speaks to falling token costs. As OpenAI and Anthropic have cut prices, average token costs have declined to $0.68 per million tokens, as opposed to the 2026 peak of $1.15 per million tokens in March.

The data suggests that the labs have yet to make up for the price cuts with growing volume. And the same incentives have many customers choosing to use older, cheaper models like OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet instead of the more powerful frontier releases. Employees at frontier labs have said much of the cost of training is recouped in the first weeks of a new model’s release, and slower adoption could threaten that dynamic.

Still, for all the talk of open-weight models threatening the frontier labs, only 6.4% of AI-spending businesses used model-serving or inference platforms in August — a share that’s growing steadily but not fast enough to drive the dynamics of broader business adoption.

“We are showing that competition between OpenAI and Anthropic is making AI more accessible, and also driving the price down for companies — and not just driving the price down, but driving spend down at the top 1% of companies that previously the market was expecting to drive much of the growth going forward,” Kharazian said.

That also helps explain the focus at AI labs on winning over nontechnical users for AI co-working tools.

This data point — dare we call it a blip? — could be a bad sign if you’re a model builder or a hyperscaler with a couple hundred billion of chips on order. But, Kharazian notes, “it depends on who you are in the market. If your company is using AI, it’s great.”

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Tim Fernholz is a journalist who writes about technology, finance and public policy. He has closely covered the rise of the private space industry and is the author of Rocket Billionaires: Elon Musk, Jeff Bezos and the New Space Race. Formerly, he was a senior reporter at Quartz, the global business news site, for more than a decade, and began his career as a political reporter in Washington, D.C.

You can contact or verify outreach from Tim by emailing tim.fernholz@techcrunch.com or via an encrypted message to tim_fernholz.21 on Signal.

View Bio