Companies are using more AI than ever but are not spending more on it. A new report from Ramp, the corporate card and expense platform, shows token spending across its customer base grew only 1.5 percent in total across August and September, after growing 42 percent in July alone. Token volumes rose through the same period, which means the average price customers pay per unit of AI work fell sharply.
Ramp lead economist Ara Kharazian told TechCrunch that almost all AI spending in the dataset comes from venture backed, high growth technology companies, and that the heaviest users have started to pull back. The top 1 percent of firms, which had been spending just under 10,000 dollars per employee per month at their summer peaks, slipped below that level in August and September, a stutter Kharazian said is sufficient to slow overall revenue growth despite rising volumes. Median firms spend only 12 to 15 dollars per employee per month, and the top 10 percent spend 600 to 1,000 dollars, so growth in the middle of the distribution does not offset a plateau at the top.
The mechanism is price, not demand. OpenAI halved API prices for its GPT-6 Sol and GPT-6 Luna models in September, and Anthropic cut prices for its Fable 5.1 and Opus 5.5 models in response, a move the company said in a recent filing reduced its revenue growth by about 600 million dollars. Enterprise discounts intended to lock in spending are also weighing on realised revenue, while customers route simple tasks to cheaper models, including open models from Meta, Nvidia backed offerings and Chinese providers, and adopt budget controls and model routers as their usage matures.
Spending did re-accelerate in the most recent window, with token spend up 8.9 percent from September 23 to October 23, but Kharazian said the headwinds to spend growth are strong and that the era of hypergrowth in AI spending may be over. The read matters beyond the model companies themselves: the case for the current scale of data center and chip spending rests on enterprise AI revenue compounding at exceptional rates, and Ramp’s data is among the first large transaction datasets to show usage and spending decoupling.