A new report from Ramp finds that companies are using more AI than before, but their spending on AI flattened in late summer: token-related spending rose by only 1.5% across August and September. By contrast, token spend grew 42% from June to July.
Why this matters for major model providers
Ramp’s data indicate that price competition and more frugal customer behavior may be capping revenue for large model providers such as OpenAI and Anthropic. Ramp’s lead economist Ara Kharazian told TechCrunch that “almost the entire spend on AI is from venture-backed, high-growth tech companies. 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 he sees the top 1% of firms — the most AI-intensive users — modestly pulling back in August and September, failing to sustain the roughly $10,000 per-employee peaks they hit over the summer. A slowdown among that small cohort, he argues, is sufficient to slow provider revenue growth even as token volumes increase.
Who is spending what?
Ramp shows growth among median firms and the top 10%, but those gains do not offset the softening at the very top. Median companies spend about $12–$15 per employee per month, while the top 10% spend $600–$1,000 per employee per month.
Kharazian noted that current use cases don’t guarantee broad adoption up to the roughly $10,000-per-employee-per-month level seen among the top 1%. While there is strong demand for AI coding agents in sectors like software and finance, other businesses — for example a doctor’s office or a car dealership — may not require that level of AI expenditure.
Price cuts and competition have shifted power to buyers
Kharazian emphasized the slowdown is not a demand problem — volumes are up — but a competitive-market effect. OpenAI and Anthropic have cut prices, making AI cheaper to use. In September, OpenAI cut API prices in half for GPT-6 Sol and GPT-6 Luna. Anthropic also made significant price reductions for Fable 5.1 and Opus 5.5; according to a filing cited in the report, those cuts reduced its revenue growth by about $600 million. Both companies have offered enterprise discounts intended to lock in spending, yet these measures still pressure revenue targets.
This dynamic, Kharazian said, has made it a buyer’s market. Big players like Meta and Nvidia have introduced enterprise offerings with low-cost or open-source models, and there are many other open-model companies from China and the U.S. A new entrant called Jev — a lightweight tool for selecting among a few predetermined outputs — is climbing Ramp’s rankings as a top performer.
“For a lot of AI use cases, you just need a simple Yes/No,” Kharazian said. Moving workloads to cheaper tools like Jev can further reduce already low costs.
Greater efficiency also reduces spend
Another reason spending is not rising in step with usage is that firms are becoming better at economizing AI use. Discounts from providers play a role, but so does improved internal control: companies increasingly use defaults that avoid routing long-running tasks to very expensive models, set spending budgets for employees, and offload work from standard models to cheaper light models via routing.
Notion co-founder Akshay Kothari shared internal data that echoed Ramp’s findings, summing it up: “tokens go brr. Costs go ahhh.” (Tweet date: October 22, 2026.)
Market consequences and investor concerns
Bloomberg reported that Anthropic’s potential valuation has dropped from an early-October estimate of $2 trillion to roughly $1.2–$1.3 trillion, and that credit agencies may give the company a high “junk” rating. OpenAI reportedly trimmed its proposed valuation from $1.4 trillion to $1.25 trillion and reduced its latest fundraising targets by about $10 billion.
Kharazian said Ramp’s dataset likely informs many investor worries. One possible upside — though speculative — is that future commercially productive use cases could trigger a new wave of spending. AI coding has been the major commercially successful application over the past year and a half, helping drive trillion-dollar valuations for companies like Anthropic and OpenAI.
Some observers expect personal agents to be the next big product. OpenAI, SpaceXAI, and Meta have released personal agents, and startups such as Instinct are pursuing similar offerings. Kharazian remains skeptical that personal agents, which he believes are “not as economically productive as coding,” will deliver the revenue boost required for uninterrupted hypergrowth.
Conclusion
Ramp’s data show that while token volumes have risen, spending growth on AI stalled in late summer — and though a re-acceleration was seen in October (token spend rose 8.9% from September 23 to October 23, per Ramp), price cuts, heightened competition and improved user-side efficiency are creating strong headwinds for revenue growth among model providers. The trend raises new questions about the long-term revenue trajectories of major AI companies and about investor expectations for perpetual hypergrowth.
