More AI, flat token spending
In Ramp’s data, companies used more artificial intelligence through the late summer — and token spending barely budged.
New data from corporate card company Ramp shows that spending on AI tokens grew just 1.5% across August and September, after jumping 42% in July — even as the volume of tokens consumed kept climbing. The figures suggest two forces at work — customers getting better at controlling what they pay, and sellers cutting prices — though the data does not settle which mattered more.
It is the kind of number that would be a footnote in any other market. In a market where ever-rising spending has been the working assumption, two flat months stand out. Ramp’s lead economist, Ara Kharazian, put it bluntly: “The era of hypergrowth may be over, and now that we know there may not be as much juice to squeeze out of the top 1%.”
The top 1% is where the story lives. These are the most “AI-pilled” firms, the heaviest users, and they stuttered slightly in August and September, landing just below the roughly $10,000 per employee per month they reached at their summer peaks. In September they spent $9,568 per employee — up 11.7% from August, but still 3.1% below July’s peak of $9,873. The median firm, by contrast, spends $12 to $15 per employee per month on AI tokens; the top 10% spend $600 to $1,000.
That concentration is what turns a wobble into a market event. One percent of companies account for 80% of business spending on OpenAI and Anthropic, according to Ramp. “Almost the entire spend on AI is from venture-backed, high-growth tech companies,” Kharazian told TechCrunch. “That’s fine as long as AI spend continues to rise, but if it starts to slow down because of something in the market or some other factor, they will all start to slow down simultaneously, and that could lead to other cascading effects.”
The 1% appear to have learned to shop
Kharazian says the slowdown is not a demand problem: token volumes kept rising. What the data suggests changed is how the heaviest users manage their spending — and how cheap the product got. Whether thrift or price cuts did more of the work is still unresolved.
As companies get advanced in using AI, they get better at controlling it. “You’re more likely to use defaults that prevent usage of very expensive models for long-running tasks,” Kharazian said. “You’re more likely to impose some budgets on employees. You’re more likely to offload some of the spend from standard and light models to even cheaper light models via routers.”
Lightweight tools that pick the cheapest model for the job are part of the same shift. Jev, a tool for choosing between a few predetermined outputs, is increasingly among the top performers on Ramp. Kharazian notes that for many AI use cases “you just need a simple Yes/No” — so work that is already cheap can be shifted to even cheaper options.
Notion co-founder Akshay Kothari recently posted internal data that matched Ramp’s insights. As he put it: “tokens go brrr. costs go ahhh.”
The sellers cut their own prices
The customers did not do this alone. The sellers helped — by cutting prices to win and keep business, then watching the cuts eat their own revenue.
OpenAI halved API prices in September for GPT-6 Sol and GPT-6 Luna. Anthropic cut pricing for Fable 5.1 and Opus 5.5 — a decision the company said in a recent filing slashed revenue growth by about $600 million. Both companies have also been offering enterprise discounts intended to lock in spend.
The result, according to Ramp’s AI Index, is a market where the price of the product is falling almost as fast as usage is rising. The blended price per million tokens across OpenAI and Anthropic fell 26.8% to $0.56 in the four weeks to October 23. OpenAI’s fell 42.4% to $0.46; Anthropic’s fell 15.7% to $0.61. Over the same four weeks, token volume jumped 37.1% while spending rose just 8.9%. The figures do not by themselves establish a common measurement basis, and the reported data does not explain why they do not reconcile into a single total.
Kharazian points to price competition among the labs as one driver of the cuts. They could also be a land-grab: cheaper AI might expand usage and create later monetization opportunities, and the evidence does not settle which story wins.
There is also a share shift — measured during October’s rebound, not the late-summer stall. Anthropic’s share of the token spending Ramp tracks rose from 51.3% to 60.5% in the four weeks to October 23, while OpenAI’s fell from 44.2% to 35.1%. Anthropic is winning a larger slice while overall token spending grew 8.9% over the same weeks.
An IPO priced on the old story
All of this lands just as Anthropic heads for the public markets.
Bloomberg reported on November 6 that prospective IPO buyers are discussing a valuation of $1.2 trillion to $1.3 trillion, down from about $2 trillion estimated in early October, and that the company is reportedly expected to receive a high “junk” rating from credit agencies. Anthropic filed publicly to list on the Nasdaq under the ticker ANT on October 26, and filed an amended prospectus on November 2 that left its offering price range blank. Its last confirmed private valuation was $965 billion, set by a $65 billion funding round in May.
None of this is a verdict on the businesses. Anthropic reported third-quarter revenue of about $17.0 billion in its amended filing, and $33.2 billion for the first nine months of 2026, against $2.5 billion a year earlier. The question the spending data raises is whether revenue growth will keep pace with the expectations behind the reported valuations. Ramp’s two-month snapshot raises that question; it does not settle it, and no final Anthropic offering valuation has been set.
Two months is not a trend
Before declaring the hypergrowth era over, note the rebound. Token spending rose 8.9% from September 23 to October 23, and the gap between usage growth and spending growth is the story, not a collapse in demand. Token volumes kept rising while spending grew much more slowly.
There are also limits to what the data can say. Ramp notes that its Token Spend Management sample is not representative of its total AI spending volume, though the results are directionally similar. And the stall’s cause is still unresolved: how much of it is price cuts, and how much is customer thrift, is an open question the data does not yet settle.
The industry’s hope is that a new use case arrives to restart the spending engine. Some think personal agents could be the next one — but whether they will prove as economically productive as coding agents is an open question the evidence leaves unresolved.
“A lot of the assumptions that we had about the trajectory of AI spend are starting to break down,” Kharazian said. “You can use more AI, but use it for less. That’s great for the customers… but it’s not necessarily great for the model companies. And it does introduce new questions around their revenue trajectories.”
For now, Ramp’s two-month snapshot says something narrower and more interesting than a bust: the biggest spenders appear to have learned to use more AI for less, with the sellers’ price cuts as a possible co-driver — though which force mattered more is still unresolved. Whether October’s rebound holds — and whether the squeeze lasts — is still an open question.
