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

Payments company Ramp, which tracks spending at roughly 70,000 firms, reported that the share of its customers paying for AI products in August nudged up to 56%, a 0.4 percentage-point rise from July. Ramp’s data have flagged similar late-summer slowdowns in the past: its AI index last year showed little growth from August through October before activity picked up again later in the year. Ramp’s customer base skews technical, so its figures may overstate adoption compared with broader surveys — for example, an ongoing U.S. Census Bureau survey updated August 23 found only about 22% of businesses reporting AI use.

The August lull coincided with a noticeable drop in AI spend per employee among the biggest spenders. Ramp economist Ara Kharazian highlighted that the top 1% of firms in the sample cut AI spend per employee by nearly 10% to $7,205. Part of that shift likely reflects seasonal factors — many companies operate at reduced capacity in August — but it also tracks with falling token prices and customers migrating to cheaper models.

Average token costs have slid to roughly $0.68 per million tokens after peaking at about $1.15 per million tokens in March 2026, according to Ramp’s data. The price competition between providers such as OpenAI and Anthropic has encouraged many customers to rely on older, lower-cost models — cited examples include OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet — rather than the newest frontier releases. Ramp’s data indicate that the volume gains from lower prices have not yet compensated for the revenue lost to those cuts among the biggest spenders.

Other signals point to a gradual, not immediate, diffusion of on-premise or third-party inference platforms: only 6.4% of AI-spending businesses in Ramp’s sample used model-serving or inference platforms in August. For model builders and hyperscalers that have committed large hardware orders, even a modest slowdown in adoption could complicate the economics of new model launches. At the same time, lower prices and more accessible models can broaden AI use for many companies, which is one reason labs are increasingly focused on non-technical users and enterprise co-working tools.