Business adoption of AI tools nearly stalled in August, with just 56% of customers at payments company Ramp paying for AI products that month, according to spending data released September 9 covering 70,000 companies. The figure climbed only 0.4% from July, marking the second consecutive year that Ramp's metrics have shown adoption slowing between August and October. The slowdown arrives as massive infrastructure investments by frontier labs and hyperscalers depend on continued revenue growth to justify their spending.
The most dramatic shift appeared among heavy users: AI expenditure per worker at the top 1% of AI-adopting firms in the sample dropped nearly 10% to $7,205 in August. Meanwhile, average token prices fell to $0.68 per million tokens from a 2026 peak of $1.15 per million tokens in March, following price cuts by OpenAI and Anthropic. The data indicates labs haven't compensated for lower prices with higher volume. Many customers are opting for older, less expensive models like OpenAI's ChatGPT 5.6-Terra and Anthropic's Sonnet rather than premium frontier releases. Open-weight models remain a small factor: only 6.4% of AI-spending businesses used model-serving or inference platforms in August, a share growing steadily but not quickly enough to reshape broader adoption patterns.
Ramp economist Ara Kharazian noted the figures show "competition between OpenAI and Anthropic is making AI more accessible" while also "driving spend down at the top 1% of companies that previously the market was expecting to drive much of the growth going forward." The report acknowledges Ramp's metrics may overstate overall adoption due to the company's tech-focused client base—an ongoing U.S. Census Bureau survey updated August 23 shows just 22% of businesses report using AI. Still, Ramp's dataset offers one of the few direct spending indicators available and potentially serves as a leading signal for the broader market.
The August slump may reflect seasonal factors, as much of the industry takes vacation during that month, which could explain both the adoption plateau and reduced token spending. But the decline in per-employee spending at top firms also points to falling token costs eroding revenue even as usage climbs, particularly following steep price cuts earlier this year. Employees at frontier labs have said much of training costs are recouped in the first weeks after a new model launches, and slower adoption could threaten that recovery timeline. The rapid pace of AI infrastructure buildout means even modest slowdowns raise concerns, since the gobsmacking investment rests on expectations of plentiful revenue to repay it. Software engineers' adoption of agentic coding tools has driven steep usage growth thus far, but if that adoption decelerates, revenue is likely to follow.
The diverging impact depends on market position. Kharazian observed that while the data represents a warning sign for model builders and hyperscalers with hundreds of billions in chip orders, "it depends on who you are in the market. If your company is using AI, it's great." The report suggests this dynamic helps explain why AI labs are increasingly focused on winning over nontechnical users through co-working tools, seeking new revenue streams beyond the power users who are now spending less per head. Companies banking on enterprise adoption may need to recalibrate expectations if summer doldrums prove to be more than a seasonal blip. The tension between democratizing access through lower prices and maintaining the revenue growth needed to fund frontier research will likely define competitive strategy as labs race to expand their user base beyond early adopters.