AI in Business: Record Usage, Declining Spending in the United States

Aggregated transaction data from Ramp shows a decline in AI spending among American companies, even as usage reaches a record high at the end of September. The measure focuses on APIs and favors large clients, and the ongoing decline has lasted unusually long. OpenAI and Anthropic account for the majority of token spending, while open source remains a minority.

Data from 70,000 companies, limited to APIs

The Ramp AI Index relies on transaction data from over 70,000 American companies. Information regarding tokens comes from a subsample. API-related spending has been analyzed, which presents a bias in favor of large AI clients. Spending peaked in July, and while this is not the first observed decline, it is notable for its unusual duration.

Less spending, more usage since July

American companies are spending less on AI, according to Ara Kharazian, an economist at Ramp. He attributes this decline to falling prices of high-end models and the rise of standard and lightweight models, which are less expensive and more efficient. Kharazian notes that the decrease in spending comes almost entirely from competition between OpenAI and Anthropic. Meanwhile, AI usage has increased by about 50% since the spending peak in July and reached a record level at the end of September.

Anthropic and OpenAI dominate, open source under 5%

According to data from the last week of September, 51% of token spending went to Anthropic and 44.5% to OpenAI. Companies still allocate less than 5% of their spending to open source models.