Business adoption of AI tools experienced a slowdown in August, based on spending data from 70,000 companies tracked by the payment platform Ramp. The latest figures reveal that 56% of Ramp’s customers paid for AI products during the month, marking a modest 0.4% increase from July. This isn’t the first time Ramp’s metrics have flagged a deceleration—last year, their AI index showed little to no growth between August and October, only to rebound as the year wound down.

Still, given the breakneck speed of the AI expansion, even minor dips can raise red flags. The massive investments in AI infrastructure by frontier labs and hyperscalers hinge on the expectation of substantial future revenue to justify those costs. Up to now, usage has climbed sharply, especially as software engineers embraced agentic coding tools—but if that adoption momentum stalls, revenue could follow suit.

Ramp’s numbers might overstate the broader picture, though, given its tech-heavy customer base. An ongoing U.S. Census Bureau survey, updated on August 23, shows only 22% of businesses report using AI. While Ramp’s data isn’t a perfect market representation, it stands out as one of the few direct spending datasets available and could serve as an early indicator. Of course, the August timing—when much of the industry is on vacation—might explain the lull. But there are other cautionary signals for companies reliant on token spend, according to Ramp economist Ara Kharazian.

One notable trend: a significant drop in AI spending per employee among the top 1% of firms in his sample, falling nearly 10% to $7,205. That could reflect vacation-related token usage, but it also points to declining token costs. As OpenAI and Anthropic have slashed prices, the average cost per million tokens has dropped to $0.68, down from a March peak of $1.15 per million tokens.

The data suggests that labs have yet to offset these price cuts with higher volume. Meanwhile, the same cost pressures are pushing many customers toward older, cheaper models—like OpenAI’s ChatGPT 5.6-Terra or Anthropic’s Sonnet—rather than the latest frontier releases. Employees at frontier labs have noted that much of a new model’s training cost is recouped within the first weeks of release, so slower adoption could upend that model.

Despite all the chatter about open-weight models threatening the frontier labs, only 6.4% of AI-spending businesses used model-serving or inference platforms in August. That share is growing steadily but not quickly enough to reshape broader adoption dynamics. “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 dynamic also helps explain why AI labs are increasingly focused on winning over non-technical users for AI co-working tools. This data point—dare we call it a blip—could be troubling if you’re a model-builder or a hyperscaler with hundreds of billions in chips on order. But, as Kharazian notes, “it depends on who you are in the market. If your company is using AI, it’s great.”