Corporate AI spending slows despite lower token costs

Data from payments firm Ramp indicates that 56 percent of its customers paid for AI products in August, a marginal increase of just 0.4 percent from the previous month. While this maintains a high adoption rate, the momentum has visibly cooled compared to the rapid expansion seen earlier in the year.

This slowdown occurs against a backdrop of massive infrastructure investments by major technology companies. The strategy relies on the assumption that usage will grow steeply enough to generate revenue that recoups the cost of building data centers and training models. If adoption stalls, the financial model supporting these expenditures becomes precarious.

Price cuts fail to drive volume

Ara Kharazian, an economist at Ramp, points to a nearly 10 percent drop in AI spending per employee among the top one percent of firms. While some of this may be attributed to summer vacations, it also reflects a significant decline in token costs.

Average costs have fallen to $0.68 per million tokens, down from a peak of $1.15 in March. Despite these lower prices, companies have not increased their usage volume proportionally. Instead, many are opting for older, cheaper models rather than the latest, more powerful releases from labs like OpenAI and Anthropic.

Limited threat from open models

Contrary to fears that open-weight models will quickly displace proprietary services, their impact remains limited. Only 6.4 percent of businesses using AI spent on model-serving platforms in August. This share is growing, but not fast enough to fundamentally alter the current market dynamics or drive widespread adoption.

The competition between major providers is making AI more accessible and driving down prices for enterprises. However, this is also reducing total spend at the highest tier of users, who were previously expected to be the primary engine of growth.

Context for business decision makers

It is important to note that Ramp’s data may not reflect the entire market. A recent US Census Bureau survey suggests that only 22 percent of businesses report using AI, compared to the 56 percent seen in Ramp’s tech-heavy client base. Nevertheless, as a direct measure of spending, it serves as a useful leading indicator.

For companies already using AI, the lower prices are beneficial. However, for model builders and hyperscalers with hundreds of billions of dollars in chip orders, this spending slump could signal a more serious challenge to their return on investment.