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

The adoption of AI tools by businesses slowed successful August, according to spending information astatine 70,000 companies collected by the payments institution Ramp. The latest survey shows 56% of Ramp customers paid for AI products successful August, rising conscionable 0.4% from the period before.

This isn’t the archetypal clip Ramp’s metrics person shown adoption slowing down. Last year, the company’s AI scale showed small to nary maturation successful adoption betwixt August and October, lone to person maturation prime up again arsenic the twelvemonth finished.

Still, the utmost gait of the AI buildout means adjacent tiny slowdowns tin beryllium origin for concern. The gobsmacking concern successful AI infrastructure by frontier labs and hyperscalers rests connected the anticipation that determination is plenty of revenue retired determination to wage it back. Thus far, usage has grown steeply, peculiarly arsenic bundle engineers adopted agentic coding tools — but if that adoption slows down, gross is apt to dilatory arsenic well.

Ramp’s figures whitethorn overstate wide adoption, acknowledgment to the company’s techy clientele: An ongoing US Census Bureau survey of AI adoption updated connected August 23 shows conscionable 22% of businesses study utilizing AI. Ramp’s survey isn’t needfully typical of the market, but it’s 1 of the fewer nonstop spending information sets disposable and perchance a starring indicator.

To beryllium sure, this information is from August, erstwhile overmuch of the manufacture is connected vacation. That whitethorn explicate the doldrums. But determination are different informing signs for companies that beryllium connected token spend, per Ramp economist Ara Kharazian.

First, a large diminution successful AI walk per worker successful the apical 1% of firms successful his sample, falling astir 10% to $7,205. That whitethorn beryllium the vacation-token factor, but it besides speaks to falling token costs. As OpenAI and Anthropic person chopped prices, mean token costs person declined to $0.68 per cardinal tokens, arsenic opposed to the 2026 highest of $1.15 per cardinal tokens successful March.

Image Credits:Ramp / Ramp

The information suggests that the labs person yet to marque up for the terms cuts with increasing volume. And the aforesaid incentives person galore customers choosing to usage older, cheaper models similar OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet alternatively of the much almighty frontier releases. Employees astatine frontier labs person said overmuch of the outgo of grooming is recouped successful the archetypal weeks of a caller model’s release, and slower adoption could endanger that dynamic.

Still, for each the speech of open-weight models threatening the frontier labs, lone 6.4% of AI-spending businesses utilized model-serving oregon inference platforms successful August; a stock that’s increasing steadily but not accelerated capable to thrust the dynamics of broader concern adoption.

“We are showing that contention betwixt OpenAI and Anthropic is making AI much accessible, and besides driving the terms down for companies—and not conscionable driving the terms down, but driving walk down astatine the apical 1% of companies that antecedently the marketplace was expecting to thrust overmuch of the maturation going forward,” Kharazian said.

That besides helps explicate the absorption astatine AI labs connected winning implicit non-technical users for AI co-working tools.

This information constituent — situation we telephone it a blip? — could beryllium a atrocious motion if you’re a model-builder oregon a hyperscaler with a mates 100 cardinal of chips connected order. But, Kharazian notes, “it depends connected who you are successful the market. If your institution is utilizing AI, it’s great.”