AI Spending Has Stalled: Companies Use More Tokens, Pay the Same
Companies are using more AI than ever, but they are not spending more on it. That is the headline from new Ramp data reported by TechCrunch on 6 November 2026: corporate spending on AI tokens grew 42% in July, then just 1.5% across August and September combined, even as token volumes kept rising. October brought a partial recovery, with token spend up 8.9% from 23 September to 23 October, but Ramp's lead economist Ara Kharazian says the headwinds are strong. The era of hypergrowth, he says, may be over.
The numbers in one place
Token spend is the money businesses pay for the AI tokens they consume: the metered units that large language models charge for processing prompts and generating answers. Ramp tracks that spending through the corporate cards and payments of the businesses on its platform, a real-world view of what companies actually pay, rather than what AI vendors report as revenue.
The pattern is stark. July was a boom month, with token spend up 42% on the previous month. Then, across August and September together, growth was only 1.5%, flat for two months. October re-accelerated, with spend up 8.9% from 23 September to 23 October, but Kharazian's reading is that the pressure has not gone away. These are Ramp's own figures, not OpenAI's or Anthropic's revenue.
| Period | Spend change | What was happening |
|---|---|---|
| July 2026 | Up 42% month on month | Rapid growth |
| August to September 2026 | Up 1.5% across both months | Spending flatlined |
| 23 September to 23 October 2026 | Up 8.9% | Partial recovery |
Why spending stopped rising
The first thing to be clear about is what this is not. Kharazian, who publishes a series of blogs called "Cracks in the AI Thesis", told TechCrunch that the slowdown is not a demand problem: AI volumes are up, and companies are sending more tokens through the models, not fewer. The slowdown is a price problem, and a behaviour problem. As Kharazian put it: "You can use more AI, but use it for less."
The price side is the most direct. In September, OpenAI cut API prices in half for two of its GPT-6 models, GPT-6 Sol and GPT-6 Luna. Anthropic answered with significant reductions of its own, cutting prices for Fable 5.1 and Opus 5.5, and said in a recent filing that the decision cut its revenue growth by about 600 million dollars. Both companies have also been offering enterprise discounts designed to lock in spending. As Kharazian put it: "OpenAI and Anthropic price cuts have just made it cheaper to use AI."
The behaviour side is about customers getting smarter. "As you get really advanced in using AI, you get better at figuring out how to control it," Kharazian said. Companies use defaults that stop very expensive models being used for long-running tasks, impose budgets on employees, and route work to cheaper models through routers. Cheaper alternatives are also eating into frontier-model spending: Meta and Nvidia are introducing enterprise offerings with cheap or open-source models, and a new company called Jev, a lightweight tool for choosing between a few predetermined outputs, is increasingly among the top performers on Ramp.
The top 1% problem
The second crack in the thesis is concentration. Almost the entire spend on AI comes from venture-backed, high-growth tech companies, Kharazian told TechCrunch. That is fine while spending keeps rising, but if it slows, those companies will all slow down at the same time.
The concentration shows up per employee. The median firm spends just 12 to 15 dollars per employee per month on AI tokens. The top 10% spend 600 to 1,000 dollars. The top 1%, the most AI-heavy firms, were spending just below 10,000 dollars per employee per month at their summer peak, and dipped slightly in August and September. Kharazian said a slowdown among that cohort alone is "sufficient to slow down revenue entirely, despite the fact that we're seeing rising volumes".
| Group of firms | Spend per employee per month |
|---|---|
| Median firm | 12 to 15 dollars |
| Top 10% | 600 to 1,000 dollars |
| Top 1% | Just below 10,000 dollars at the summer peak, then dipping |
Kharazian doubts the rest of the economy will ever catch up to the top 1%. "If you look at the use cases today, it's not necessarily clear they will get up to that 10,000 dollars per employee per month mark that the 1% is at," he said. There is plenty of demand for AI coding agents in software and finance, but "it's not necessarily true that a doctor's office or a car dealership will need to spend that much on AI." His October update of the Ramp AI Index adds detail: the top 1% spent 9,568 dollars per employee per month in September, down 3.1% from a July high of 9,873 dollars, while token volumes rose. "It's clear this effect is not seasonal," he wrote.
What it means for Anthropic's IPO
The timing matters: Anthropic is heading for the public markets. Our earlier coverage explained how Anthropic files for its IPO, and the Ramp numbers feed into the question investors will ask: can AI companies keep growing revenue at the pace their valuations assume? Bloomberg reported that Anthropic's potential valuation has sunk from the 2 trillion dollars estimated in early October to just 1.2 to 1.3 trillion dollars, and the company is reportedly expected to receive a high junk rating from credit agencies. OpenAI lowered its proposed valuation to 1.25 trillion dollars from 1.4 trillion dollars and cut its latest fundraising goals by about 10 billion dollars. OpenAI's $11 billion first close still valued it at 1.25 trillion dollars, which shows how much money is chasing the sector even as the growth story gets harder to defend.
That tension is the heart of the debate about whether the AI boom is sustainable. Our earlier analysis asked whether the is the AI bubble bursting, and the Ramp data is the strongest evidence yet that revenue growth is not guaranteed by usage growth.
What to watch next
The hopeful case is that a new use case will restart spending. AI coding emerged only about a year and a half ago as the most commercially viable product, and it is what made Anthropic and OpenAI into trillion-dollar companies. Investors hope another revolution will drive further spend, and some think personal agents will be the next coding agents. OpenAI, SpaceXAI, and Meta have all released their own personal agents, alongside startups like Instinct. Kharazian is sceptical: personal agents, he said, are "not as economically productive as coding".
The Microsoft vs Google vs Amazon cloud earnings will be the next big test of the same question: whether the cloud giants, who sell the computing power behind the models, can keep growing while the model companies cut prices. For now, the message is clear. AI is getting cheaper, the tools for controlling costs are getting better, and the companies that route work to the cheapest model that can do the job will keep more of the benefit. The model companies have to find growth in a market where their best customers are learning to spend less.
Key Takeaways
- Ramp data shows corporate AI token spending grew 42% in July 2026, then only 1.5% across August and September, even though token volumes kept rising.
- October re-accelerated, with token spend up 8.9% from 23 September to 23 October, but Kharazian says the headwinds are strong.
- The slowdown is not a demand problem: it is falling prices, with OpenAI halving prices on two GPT-6 models and Anthropic cutting prices for Fable 5.1 and Opus 5.5, plus smarter cost control.
- Spending is highly concentrated: the median firm pays 12 to 15 dollars per employee per month, the top 10% pay 600 to 1,000 dollars, and the top 1% peaked just below 10,000 dollars before dipping.
- The figures matter for Anthropic's IPO, with Bloomberg reporting its potential valuation down from 2 trillion dollars in early October to 1.2 to 1.3 trillion dollars.
- For customers, cheaper AI is good news; for model companies, it raises questions about whether revenue can keep growing at the pace their valuations assume.
Conclusion
The Ramp data does not say AI is failing. It says AI is getting cheaper faster than companies are finding new ways to spend on it. Usage is up, and the tools are spreading into more businesses every month. But the bill is flat, the heaviest spenders are pulling back, and the price war between OpenAI and Anthropic is transferring value from sellers to buyers. That is a healthy market for everyone who buys AI. For the companies selling it, and for the investors who priced them as if hypergrowth would never end, it is the first serious warning that the easy growth may already be behind them.
Comments
Post a Comment