Companies are using more AI than ever, but they are not spending more on it. Ramp’s data shows AI spending flattened in late summer even as token volumes rose, and the pattern points to price competition and more cost-aware customers limiting revenue growth for OpenAI and Anthropic.
The spending pattern in Ramp data
Ramp reported that token spend rose only 1.5% across August and September. In July, it increased 42% from the previous month. The slowdown came even as companies kept using AI at higher volumes, which is why the report raises questions about the pace of revenue growth for model providers.
Ara Kharazian, Ramp’s lead economist, said almost all AI spending comes from venture-backed, high-growth tech companies. He said that works as long as AI spend keeps rising, but if spending slows because of market conditions or another factor, those firms could slow down together and trigger cascading effects. He also said the top 1% of firms, which he described as the most AI-pilled, had a slight stutter in August and September and landed just below the $10,000 per employee level they reached at summer peaks.
The median and top 10% still spend more, but the numbers do not close the gap. Median firms spend $12 to $15 per employee per month, while the top 10% spend $600 to $1,000. Kharazian said the current use cases do not make it clear that the 1% will reach that $10,000 per employee per month level again, and he pointed to software and finance as industries with strong demand for AI coding agents.
Pricing pressure is changing the market
Kharazian said the slowdown is not a demand problem because AI volumes are still rising. He said the main pressure comes from competition, with OpenAI and Anthropic cutting prices and making AI cheaper to use. In September, OpenAI cut API prices in half for GPT-6 Sol and GPT-6 Luna. Anthropic also reduced pricing for Fable 5.1 and Opus 5.5, and the company said in a recent filing that the move cut revenue growth by about $600 million.
Both companies have also offered enterprise discounts meant to lock in spending. Kharazian said the market has become a buyer’s market. He also argued that the old Big Tech pattern of subsidizing expensive products to keep customers does not translate well to AI because competition is too intense.
He pointed to Meta and Nvidia, which are offering enterprise products with cheap or open source models, and to other open model companies in China and the United States. He also cited Jev, a lightweight tool for choosing between a few predetermined outputs, and said it is increasingly among Ramp’s top performers. For many AI use cases, he said, a simple yes or no is enough, so companies can move from already cheap models to Jev and make the work even cheaper.
Why costs are falling despite more usage
Kharazian said companies are also getting better at controlling how they use AI. He said advanced users are more likely to use defaults that block very expensive models for long-running tasks, set budgets for employees, and shift spending from standard and light models to cheaper light models through routers. That ability to complete work at a lower cost is another reason spend growth has slowed.
Notion co-founder Akshay Kothari posted internal data that matched Ramp’s findings. He summed it up by writing, “tokens go brrr. costs go ahhh.” Kharazian said those assumptions about AI spending are starting to break down. He said customers can use more AI while spending less on it, which helps customers but complicates revenue prospects for model companies.
Valuations and investor expectations
Bloomberg reported Friday that Anthropic’s potential valuation fell from the $2 trillion figure estimated in early October to $1.2 trillion to $1.3 trillion. The company is also reportedly expected to receive a high junk rating from credit agencies. OpenAI also lowered its proposed valuation to $1.25 trillion from $1.4 trillion and cut its latest fundraising goals by about $10 billion.
Kharazian said Ramp’s dataset likely feeds investor concerns. He said one speculative bright spot remains the idea that AI will create new use cases that revive spending growth. AI coding became the most commercially viable product about a year and a half ago, and that helped turn Anthropic and OpenAI into trillion-dollar companies. Some investors hope personal agents will be the next major wave.
OpenAI, SpaceXAI, and Meta have all released personal agents, and startups like Instinct have done the same. Kharazian said he is skeptical that personal agents, which he described as less economically productive than coding, will supply the growth the industry needs for nonstop expansion. He said projections still rely on new products driving further growth, but based on what is visible today, AI spend has started to slow down.
Licensed under CC BY-ND 4.0 International