What happened
- Ramp, a corporate spend management platform, published a report showing that companies’ spending on AI flattened at the end of the northern summer, even as the volume of tokens consumed kept rising. TechCrunch reported it on 6 November 2026.
- According to Ramp’s data, token spending grew only 1.5% between August and September. In July it had risen 42% over the previous month.
- Ara Kharazian, Ramp’s lead economist, told TechCrunch that almost all of the spending comes from high-growth, venture-backed technology companies.
- October showed a new acceleration: token spending rose 8.9% between 23 September and 23 October, although Kharazian warns the headwinds are strong.
Why it matters
- Anthropic is preparing to go public and the market usually takes for granted that these companies will grow at a dizzying pace. This data suggests that assumption is not safe.
- The stall does not come from a drop in demand. AI usage keeps increasing; what changes is how much each unit costs.
- If most of the spending is concentrated in a small group of companies, a slowdown in that group is enough to slow the whole sector’s revenue.
The number
US$ 10,000 per employee a month. That is the level reached, at its summer peak, by the 1% of the most AI-intensive companies, according to Ramp. In August and September they came in slightly below it. Median companies spend between US$ 12 and US$ 15 per employee a month, and the top 10% between US$ 600 and US$ 1,000.
Context
Kharazian attributes the slowdown to competition among providers. In September, OpenAI cut API prices for GPT-6 Sol and GPT-6 Luna in half, and Anthropic answered with reductions for Fable 5.1 and Opus 5.5. According to TechCrunch, a recent Anthropic filing says that decision trimmed its revenue growth by about US$ 600 million.
On top of that come the discounts for large buyers, which aim to lock in spending but end up hurting revenue targets, and the supply of cheap open models from Meta, Nvidia and companies in China and the United States. Kharazian also mentions Jev, a lightweight tool that picks among a few predefined outputs: for many tasks a yes or no is enough, and moving already cheap models to Jev makes them cheaper still.
Companies, for their part, learn to spend better: they set per-employee budgets and avoid using expensive models on long tasks. All of this comes from Ramp’s data, which reflects the companies that use that platform rather than the whole market.
At Mamífero we have followed this industry’s money. We reported that Microsoft declared US$ 10 billion of revenue from OpenAI in a single quarter and that Fal raised US$ 1.3 billion and doubled its value. Also that Legora raised US$ 550 million and doubled its annual revenue and that OpenAI opened its decisions API to everyone.
What’s next
- Anthropic’s listing will test whether investors price in slower growth.
- The price war between providers and open models will keep pressing revenue per unit.
- Ramp’s next reports will show whether the October rebound holds or was a bounce.
Bottom line
More usage does not mean more revenue when each token costs less. For AI companies, the challenge shifts from winning users to selling each query better.
Sources
- TechCrunch, “Ramp data shows AI spending stalling just as Anthropic nears its IPO”, 6 November 2026, link. Figures are Ramp’s, as cited by the outlet.
- Cover image: Lotus Head from Johannesburg, Gauteng, South Africa, CC BY-SA 2.5, via Wikimedia Commons
Edited by Rodrigo Cornejo. How we select and verify: who writes these notes.
