Corporate adoption of artificial intelligence continues to climb, but the dollars flowing to AI providers are not keeping pace. A new spending analysis from financial technology firm Ramp reveals that AI expenditure among businesses effectively stalled during late summer, even as the volume of AI usage - measured in tokens processed - continued to rise. The divergence points to a market undergoing a significant shift: customers are using AI more efficiently and paying less per unit of output, which is beginning to compress revenue growth for the major AI model providers.

According to Ramp's data, token spend grew by just 1.5% across August and September combined. That figure is a dramatic deceleration from July, when spending surged 42% compared to the prior month. The timing is notable, as Anthropic is currently preparing for an initial public offering, raising questions about whether investors have priced in realistic growth expectations for the AI sector.

A Concentrated Spending Base Starts to Wobble

Ramp's lead economist, Ara Kharazian, offered a pointed observation about who is actually driving AI spending. "Almost the entire spend on AI is from venture-backed, high-growth tech companies," he noted. "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."

Kharazian said the most telling signal came from the top 1% of AI-spending firms - the heaviest users in Ramp's dataset. These companies saw their per-employee spending dip slightly in August and September, falling just short of the $10,000 per-employee monthly threshold they had reached during the summer. That marginal pullback among the most committed spenders was, he said, enough to drag overall revenue growth into near-stagnation. "A slowdown among that cohort is sufficient to slow down revenue entirely, despite the fact that we're seeing rising volumes," he explained.

For context, Ramp's data shows the median company spends between $12 and $15 per employee per month on AI, while the top 10% of spenders fall in the $600 to $1,000 range. The gap between median and top-tier users is enormous, and while both groups showed some growth, it was not enough to offset the plateau at the very top.

Kharazian raised a broader structural question about whether most businesses will ever reach the spending levels seen among the most intensive AI users. "If you look at the use cases today, it's not necessarily clear they will get up to that $10,000 per employee per month mark that the 1% is at," he said. Demand for AI coding tools is strong in software and finance, but he noted that sectors like healthcare or automotive retail may simply not have the same need or capacity for that level of AI investment. "It's not necessarily true that a doctor's office or a car dealership will need to spend that much on AI."

Price Wars Are Cutting Into Revenue

October brought some relief, with token spending rising 8.9% between September 23 and October 23. However, Kharazian cautioned that the broader headwinds on spending growth remain considerable. He has been documenting these trends in a series of posts titled "Cracks in the AI Thesis," and he is careful to distinguish between a demand problem and a pricing problem. The data, he argues, reflects the latter.

"OpenAI and Anthropic price cuts have just made it cheaper to use AI," he said. In September, OpenAI reduced its API prices by half for its GPT-6 Sol and GPT-6 Luna models. Anthropic followed suit with significant price reductions on its Fable 5.1 and Opus 5.5 models - a move the company itself acknowledged in a regulatory filing had cut revenue growth by approximately $600 million. Both companies have also extended enterprise discounts designed to lock in long-term customers, but those arrangements are simultaneously undercutting their own revenue targets.

The result, Kharazian said, is that the market has tilted firmly in favor of buyers. Historically, large technology companies have subsidized access to expensive services to build user dependency, with the expectation of raising prices once competition thins out. The ride-hailing industry followed that playbook - early Uber and Lyft fares were heavily subsidized, and prices have risen substantially as the market consolidated. Kharazian argues that dynamic does not translate to AI, because the competitive landscape is simply too crowded and too open.

"Meta and Nvidia are also introducing enterprise offerings with pretty cheap models or open source models," he said, adding that a growing number of open-weight model providers from both China and the United States are adding further downward pressure on pricing. He also highlighted a newer entrant called Jev - described as a lightweight tool that routes queries toward a set of predetermined outputs - which has been climbing the rankings among top AI tools used by Ramp customers.

"For a lot of AI use cases, you just need a simple Yes/No," Kharazian said. "So you take your already cheap models, you shift them over to Jev, and now your already cheap stuff gets even cheaper."

Enterprises Are Getting Better at Spending Less

Beyond falling prices, companies are also becoming more disciplined and sophisticated in how they deploy AI, which is itself a factor reducing total expenditure. Kharazian described a maturation curve where more experienced AI users learn to constrain costs proactively.

"As you get really advanced in using AI, you get better at figuring out how to control it," he said. "You're more likely to use defaults that prevent usage of very expensive models for long-running tasks. 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."

This pattern was independently corroborated by Akshay Kothari, co-founder of productivity platform Notion, who recently shared internal data reflecting the same trend. In a widely circulated post, Kothari summarized the dynamic in simple terms: token usage is climbing sharply while costs are falling. The observation aligned closely with what Ramp's analysis had found across a much broader dataset.

"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."

Valuations Under Pressure as IPO Approaches

The spending data is landing at a sensitive moment for Anthropic. Reports from Bloomberg indicate that the company's estimated valuation has already declined - from roughly $2 trillion projected in early October to a revised range of $1.2 to $1.3 trillion. The company is also reportedly expected to receive a high-yield, or "junk," credit rating from major rating agencies, a designation that signals elevated financial risk to prospective investors.

OpenAI has similarly revised its financial expectations downward, reducing its proposed valuation from $1.4 trillion to $1.25 trillion and cutting its latest fundraising target by approximately $10 billion. The parallel adjustments at both companies suggest that the market is beginning to incorporate more cautious assumptions about near-term revenue growth.

Kharazian acknowledged that Ramp's dataset is likely informing some of the investor skepticism now circulating around AI valuations. The data provides a ground-level view of actual enterprise spending behavior, as opposed to projections or forward-looking guidance from the companies themselves.

The Search for the Next Growth Catalyst

There remains one argument that bulls in the AI sector continue to rely on: the possibility that new categories of AI products will emerge and drive a fresh wave of spending, much as AI coding tools did roughly eighteen months ago. The rise of coding assistants is widely credited with helping push both Anthropic and OpenAI into trillion-dollar valuation territory, and investors are hoping history will repeat itself.

The leading candidate for the next transformative category is personal AI agents - software that can autonomously carry out tasks on behalf of individual users. OpenAI, Meta, and others have already released early versions of personal agent products, and a number of startups including Instinct are competing in the space as well.

Kharazian is not convinced that personal agents will deliver the same economic lift that coding tools did. He pointed out that coding agents serve a highly productive, commercially measurable function - they help software teams ship faster and reduce engineering overhead. Personal agents, by contrast, handle tasks that are often less directly tied to revenue generation. "Not as economically productive as coding," he said of the personal agent category, suggesting its spending impact may be more modest.

"The projection is that there will be new products being introduced that will drive further growth," Kharazian said. "But based on what we can observe today and the products available and what is commercially advanced, AI spend has started to slow down."

Whether the slowdown is a temporary plateau before another acceleration or the beginning of a more sustained deceleration remains an open question. What Ramp's data makes clear is that the relationship between AI usage and AI revenue is no longer as straightforward as it once appeared - and that the assumptions underlying some of the sector's loftiest valuations may need to be revisited.