The pace of adoption of artificial intelligence tools in business slowed in August, according to data from payments company Ramp, collected among 70,000 companies.
The pace of adoption of artificial intelligence tools in business slowed in August, according to data from payments company Ramp, collected among 70,000 companies.
In August, AI-based products were paid for by 56% of Ramp customers, up just 0.4% from the month before, TechCrunch reports .
This is not the first time that this has happened: last year, Ramp also showed minimal growth from August to October, before recovering strongly towards the end of the year. However, given the massive investments in AI infrastructure by leading labs and hyperscalers, even a slight slowdown is worrying market analysts.
Ramp’s numbers may slightly overstate overall adoption due to the technology-centric nature of its customer base (only 22% of businesses report using AI, according to the U.S. Census Bureau as of August 23). However, Ramp’s analysis remains one of the few direct sources of data on actual spending and serves as a leading indicator of the market.
Ramp economist Ara Kharazyan calls the main alarming signal the significant drop in AI spending per employee among the top 1% of companies that most actively use AI — this figure decreased by almost 10%, to $7,205.
Key factors driving cost decline:
seasonality - the August holiday period traditionally reduces corporate activity;
Declining token value — due to price competition between OpenAI and Anthropic, the average token value fell to $0.68 per million tokens compared to a peak of $1.15 in March 2026.
Downsizing. Customers are increasingly choosing older or lighter models (such as ChatGPT 5.6-Terra or Anthropic Sonnet) over the most expensive flagship versions.
So far, developers have not been able to compensate for the price decline with a corresponding increase in consumption volumes. At the same time, the share of companies using platforms to deploy open models is only 6.4%, which indicates that the main competition between the main market players remains.
According to Kharazyan, fierce competition makes technologies more accessible and lowers the bill for customers. If for model developers and computing power providers such a slowdown in costs is a risk, then for end users in the business segment it is a positive trend.
