Ramp Data Shows Business AI Spending Stalled Over the Summer as Token Use Kept Rising

Token spend among Ramp's customers rose only 1.5% across August and September, Ramp lead economist Ara Kharazian told TechCrunch. In July it had grown 42% from the month before. Usage volumes kept climbing over the same period, which means customers were paying less for each unit of AI they consumed.

The gap between usage and spending goes to the heart of the investment case for the large model developers. Investors weighing Anthropic's flotation, and OpenAI's own fundraising, have largely assumed that revenue will keep compounding at the pace of the past year. Kharazian said the data points to slower growth. "A lot of the assumptions that we had about the trajectory of AI spend are starting to break down," he told the outlet. "You can use more AI, but use it for less."

Spending is highly concentrated, which makes the slowdown at the top matter more. Kharazian said almost all AI spend comes from venture-backed, high-growth technology companies, and that the top 1% of firms, which had been spending about $10,000 per employee per month at their summer peak, slipped just below that level in August and September. A slowdown in that group alone is enough to flatten overall revenue despite rising volumes, he said. Median firms spend $12 to $15 per employee per month and the top 10% spend $600 to $1,000, and while both groups are growing, Ramp's figures show it is not enough to offset the pullback at the top.

Kharazian also questioned whether other industries will ever approach the spending of the heaviest users. Coding agents are in heavy demand in software and finance, he said, but "it's not necessarily true that a doctor's office or a car dealership will need to spend that much on AI."

October brought some recovery, with token spend up 8.9% between September 23 and October 23, but Kharazian said the forces holding down spending remain strong. He attributed the slowdown to competition rather than weak demand. In September, OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna, and Anthropic followed with significant cuts to pricing for Fable 5.1 and Opus 5.5, a move TechCrunch said the company disclosed in a recent filing had reduced its revenue growth by about $600 million. Enterprise discounts intended to lock in customers are also weighing on revenue, according to the report.

Cheaper alternatives are adding to the pressure. Kharazian pointed to low-cost enterprise offerings and open models from Meta and Nvidia, open model developers in China and the United States, and a tool called Jev that lets companies send simple yes-or-no tasks to even cheaper systems. Experienced customers are also getting better at controlling costs, he said, by setting defaults that keep expensive models away from long-running tasks, putting budgets on employees and routing work to lighter models. Notion co-founder Akshay Kothari recently posted internal figures pointing the same way, summing them up as "tokens go brr. Costs go ahhh."

The data adds to pressure on valuations that has built through the week's technology sell-off. Bloomberg reported on Friday that some investors in Anthropic's planned offering are now discussing a valuation of $1.2 trillion to $1.3 trillion, down from the roughly $2 trillion floated in early October. TechCrunch also reported that OpenAI had lowered its proposed valuation to $1.25 trillion from $1.4 trillion and cut its latest fundraising target by about $10 billion.

Kharazian, who has been publishing a series of Ramp posts titled "Cracks in the AI Thesis", said the dataset probably shapes many of the concerns investors now have. The main upside for spending, in his account, is the hope that a new category of use, such as personal agents, will do for model revenue what coding tools did over the past year and a half. Ramp's next monthly reading, covering October and November, will show whether the October rebound held.