Enterprise AI spending per worker fell in August as model prices dropped, Ramp says

Enterprise spending on AI tools slowed in August, according to spending data from the payments company Ramp, which tracks card and bill payments at about 70,000 companies. Ramp said 56% of its customers paid for an AI product in August, a rise of 0.4% from July. The company’s economist, Ara Kharazian, pointed to a sharper signal underneath that flat line: among the top 1% of AI-using firms in his sample, spending per employee fell nearly 10%, to $7,205.

Ramp links the drop to falling prices for tokens, the units of text that AI models process and that labs bill for. Average token costs came down to $0.68 per million tokens in August, Ramp said, against a peak of $1.15 per million in March 2026. OpenAI and Anthropic have both cut prices this year. Ramp’s reading is that the two companies have not yet made up for those cuts with higher usage, so the customers expected to drive future growth are paying less.

The same pressure is changing which models companies pick. Ramp said many customers are choosing older, cheaper options such as OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet rather than the newest frontier releases. That matters to the labs because, as Ramp notes, employees at frontier labs have said much of a model’s training cost is recovered in the first weeks after release. Slower uptake of new models weakens that pattern.

Ramp’s numbers likely run ahead of the wider economy because its customer base skews toward technology firms. A U.S. Census Bureau survey of AI adoption, updated on August 23, found just 22% of businesses report using AI. Ramp’s survey is not built to represent the whole market, but it is one of the few direct measures of what companies actually spend, and it may move before other indicators do.

There are reasons to discount the August figures. Much of the industry is on vacation in August, which cuts token use. Ramp’s own AI index showed little to no growth between August and October last year before adoption climbed again at the end of the year. Open-weight models, which companies can run on their own hardware, are also not yet a major factor: only 6.4% of AI-spending businesses used model-serving or inference platforms in August, a share Ramp said is growing steadily but slowly.

“Competition between OpenAI and Anthropic is making AI more accessible, and also driving the price down for companies,” Kharazian said, adding that it is also “driving spend down at the top 1% of companies that previously the market was expecting to drive much of the growth going forward.” He said the effect cuts both ways: “it depends on who you are in the market. If your company is using AI, it’s great.”

The next Ramp AI index, covering September, will show whether the August dip was a vacation lull or the start of a longer flattening. Also worth watching: the Census Bureau’s next adoption update, and whether the small share of firms using inference platforms keeps climbing.