AI use rises but spending stalls: Ramp data raises concerns just as Anthropic nears its IPO

A report from Ramp, cited by TechCrunch, says companies are using more AI than ever but are not spending more on it. According to Ramp's data, token spending grew just 1.5% between August and September, compared with a 42% month-on-month increase in July. Token volumes kept rising over the same period, which suggests that price competition and more sophisticated customers may be limiting revenue at OpenAI and Anthropic.

The figures come as Anthropic prepares for its IPO, and they suggest investors may be wrong to assume AI companies will keep growing at breakneck speed. Ara Kharazian, Ramp's chief economist, explains that almost all AI spending comes from high-growth, venture-backed tech companies. As long as spending keeps rising, there is no problem. But if any market factor slows it down, all of those companies would slow down at once, with possible knock-on effects.

Kharazian already sees signs of this among the top 1% of companies most "all-in" on AI. They wavered slightly in August and September, landing just below the $10,000 per employee they reached at their summer peaks. He says a slowdown in such a small group is enough to stall revenue growth entirely, even as volumes rise. Spending is growing among median companies and the top 10%, but not enough to make up for it. The median company spends $12 to $15 per employee per month, and the top 10% spend between $600 and $1,000.

The economist doubts other sectors will reach those levels. There is strong demand for coding agents in software and finance, but it is not clear that a medical practice or a car dealership needs to spend that much on AI. "The era of hypergrowth may be over," he says. October's data showed a reacceleration, with token spending up 8.9% between September 23 and October 23, but Kharazian maintains that the headwinds for spending growth remain strong.

According to Kharazian, the cause is not a lack of demand but the competitive nature of the market. Price cuts by OpenAI and Anthropic have made AI cheaper to use. In September, OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna. Anthropic responded with significant cuts to Fable 5.1 and Opus 5.5, a move the company itself said in a recent filing reduced its revenue growth by about $600 million. Both also offer enterprise discounts meant to lock in spending, though in practice these are making it harder to hit their revenue targets. For Kharazian, it has become a buyer's market.

Big Tech's usual playbook of subsidizing a service to win customer loyalty, as Uber and Lyft did, would not work here because there is too much competition. Meta and Nvidia offer enterprise packages with cheap or open-source models, and there are other open-model companies in China and the United States. He also points to Jev, a lightweight tool for choosing among a few predefined outputs, which increasingly shows up among the top performers on Ramp. In many cases a simple yes/no is enough, so workloads that were already cheap move to Jev and get cheaper still.

Another factor is that companies are becoming more frugal. Those who have mastered AI are learning to keep it in check. They use default settings that stop very expensive models from being used on long tasks, set budgets for employees, and use routers to shift usage from standard models to lightweight ones. Akshay Kothari, co-founder of Notion, published internal data that matched these findings and summed it up this way: "tokens go brrr. costs go ahhh." Kharazian concludes that companies can use more AI while spending less. That is good for customers, but it raises questions about the revenue trajectory of the model makers.

The financial backdrop is delicate. Bloomberg reported on Friday that Anthropic's potential valuation had fallen from the $2 trillion estimated in early October to between $1.2 trillion and $1.3 trillion, and that credit agencies are expected to give it a high "junk bond" rating. OpenAI has also lowered its proposed valuation to $1.25 trillion from $1.4 trillion and cut its fundraising target by about $10 billion. Kharazian believes Ramp's data is probably feeding many of investors' concerns.

There is one possible upside, though it is speculative: the hypergrowth thesis rests on new use cases emerging. AI coding emerged a year and a half ago as the most commercially viable product and pushed Anthropic and OpenAI to trillion-dollar valuations. Some believe personal agents will be the next big driver. OpenAI, SpaceXAI and Meta have already launched theirs, as have startups such as Instinct. Kharazian is skeptical, because these agents are "less economically productive than coding." Based on what can be seen today, he concludes, AI spending has started to slow.