AI spend per worker slumped at prime corporations in August — summer season doldrums or a warning signal?

The adoption of AI instruments by companies slowed in August, in keeping with spending knowledge at 70,000 corporations collected by the funds firm Ramp. The latest survey exhibits 56% of Ramp clients paid for AI merchandise in August, rising simply 0.4% from the month earlier than.

This isn’t the primary time Ramp’s metrics have proven adoption slowing down. Final 12 months, the corporate’s AI index confirmed little to no development in adoption between August and October, solely to have development choose up once more because the 12 months completed.

Nonetheless, the intense tempo of the AI buildout means even small slowdowns will be trigger for concern. The gobsmacking funding in AI infrastructure by frontier labs and hyperscalers rests on the hope that there’s plenty of revenue on the market to pay it again. Up to now, utilization has grown steeply, significantly as software program engineers adopted agentic coding instruments — but when that adoption slows down, income is prone to sluggish as effectively.

Ramp’s figures could overstate total adoption, because of the corporate’s techy clientele: An ongoing US Census Bureau survey of AI adoption up to date on August 23 exhibits simply 22% of companies report utilizing AI. Ramp’s survey isn’t essentially consultant of the market, but it surely’s one of many few direct spending knowledge units out there and doubtlessly a number one indicator.

To make certain, this knowledge is from August, when a lot of the trade is on trip. That will clarify the doldrums. However there are different warning indicators for corporations that rely upon token spend, per Ramp economist Ara Kharazian.

First, a serious decline in AI spend per worker within the prime 1% of corporations in his pattern, falling practically 10% to $7,205. Which may be the vacation-token issue, but it surely additionally speaks to falling token prices. As OpenAI and Anthropic have lower costs, common token prices have declined to $0.68 per million tokens, versus the 2026 peak of $1.15 per million tokens in March.

The information means that the labs have but to make up for the value cuts with rising quantity. And the identical incentives have many purchasers selecting to make use of older, cheaper fashions like OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet as a substitute of the extra highly effective frontier releases. Staff at frontier labs have mentioned a lot of the price of coaching is recouped within the first weeks of a brand new mannequin’s launch, and slower adoption might threaten that dynamic.

Nonetheless, for all of the discuss of open-weight fashions threatening the frontier labs, solely 6.4% of AI-spending companies used model-serving or inference platforms in August; a share that’s rising steadily however not quick sufficient to drive the dynamics of broader enterprise adoption.

“We’re exhibiting that competitors between OpenAI and Anthropic is making AI extra accessible, and likewise driving the value down for corporations—and never simply driving the value down, however driving spend down on the prime 1% of corporations that beforehand the market was anticipating to drive a lot of the expansion going ahead,” Kharazian mentioned.

That additionally helps clarify the main focus at AI labs on winning over non-technical users for AI co-working instruments.

This knowledge level — dare we name it a blip? — may very well be a nasty signal should you’re a model-builder or a hyperscaler with a pair hundred billion of chips on order. However, Kharazian notes, “it is determined by who you’re available in the market. If your organization is utilizing AI, it’s nice.”