Business adoption of paid AI tools continued to rise in August, but new spending data suggests the pace weakened. Ramp found that 56 percent of its customers paid for AI products during the month, an increase of only 0.4 percentage points from July. Spending per employee also fell among the heaviest users.
The payments company draws its index from 70,000 businesses. That provides a direct view of purchases rather than survey intentions, but Ramp’s technology-oriented customer base may use AI more heavily than the wider economy. A US Census Bureau survey updated on August 23 found that 22 percent of businesses reported using AI.
Adoption Rose While Momentum Flattened
August was not the first weak month in Ramp’s series. Its index showed little or no adoption growth between August and October last year before activity accelerated near year-end. Summer vacations could again explain part of the slowdown, making one month’s result a possible seasonal pause rather than a structural change.
The stakes are larger because AI laboratories and cloud providers have committed heavily to infrastructure on the assumption that customer revenue will grow. Coding agents helped drive early usage, particularly among software developers. If the number of paying businesses or the amount they consume stops expanding, providers may take longer to recover investments in models and chips.
Ramp’s data cannot answer that long-term question yet. It covers its own customers, and the 0.4-point increase still represents growth. The result is best treated as an indicator to watch alongside later months rather than proof that enterprise demand has peaked.
Lower Prices Reduced Spending at the Top
The largest change appeared among the top 1 percent of companies in Ramp’s sample. Their AI spending per employee fell nearly 10 percent to $7,205. Fewer working days may have reduced usage, but lower token prices offer another explanation.
Average token cost dropped to $0.68 per million tokens in August, compared with a 2026 peak of $1.15 in March. OpenAI and Anthropic have cut prices, and customers can select older, cheaper systems such as OpenAI’s ChatGPT 5.6-Terra and Anthropic’s Sonnet instead of the newest frontier releases. Falling unit prices are good for buyers, even when they reduce a provider’s revenue from the same workload.
Ramp economist Ara Kharazian says the data suggests laboratories have not yet made up the price reductions through higher volume. TechCrunch reports that employees at frontier labs have said much of a model’s training cost can be recovered during the first weeks after release. That claim is not accompanied by audited cost or revenue figures, but it illustrates why early adoption of expensive new models matters to providers.
Open Models Have Not Yet Changed the Pattern
Only 6.4 percent of AI-spending companies in Ramp’s data used model-serving or inference platforms in August. That share is growing steadily, but not fast enough to define overall business adoption. The figures therefore do not support a conclusion that open-weight models have already displaced frontier providers across the market.
They do show two different versions of progress. More businesses are buying AI tools, while intense competition makes each unit of model use cheaper. Customers can benefit even if spending growth disappoints infrastructure investors.
The next few months will clarify whether August reflected vacations, price efficiency or weaker appetite for new tools. A rebound would repeat last year’s seasonal pattern. Continued stagnation, especially in usage volume, would challenge expectations that declining prices automatically unlock enough demand to finance the industry’s buildout. For now, the defensible conclusion is narrower: adoption edged upward, heavy-user spending fell and cheaper tokens shifted value toward enterprise customers rather than model providers.
