Ramp data suggests companies are processing more AI work without spending much more, helped by price cuts, discounts, and tighter internal controls. For employers, the next phase of AI adoption may be less about buying more tools and more about managing usage well.
Employers may be entering a more disciplined phase of AI adoption. New data from Ramp indicates that companies are using AI more heavily, but the amount they pay for that usage has stopped rising at the same pace.
For managers, that matters. The headline is not that AI demand has disappeared. It is that teams are becoming more cost conscious, vendors are cutting prices, and buyers have more options. In practical terms, AI budgets may need to be managed less like experimental innovation spend and more like everyday software infrastructure, with controls, policies, and clear ownership.
AI Use Is Rising While AI Bills Stop Climbing | Hybrion Insights
Ramp sees AI usage rising while spend levels off
Ramp’s latest AI spending data shows a sharp change from earlier growth patterns. According to the company, spending on AI tokens rose only 1.5 percent across August and September. In July, by contrast, token spending had increased 42 percent from the prior month.
Tokens are the units of text and data that AI systems process when users send prompts, receive answers, or run automated workflows. A company can process more tokens if employees use AI more often, if applications run more AI tasks in the background, or if systems handle larger workloads.
Ramp’s finding is that token volumes are still climbing, but the associated spending has flattened. Ara Kharazian, Ramp’s lead economist, told TechCrunch that AI spending is concentrated among venture-backed, fast-growing technology companies. That concentration creates risk if the highest-spending group slows at the same time.
“Almost the entire spend on AI is from venture-backed, high-growth tech companies,” Kharazian said. “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.”
The top 1 percent of AI-spending firms, which Ramp has described as the most “AI-pilled,” had been spending around $10,000 per employee per month at their summer highs. Kharazian said that group dipped slightly below that level in August and September. He argued that a pullback in this small but important cohort can be enough to slow overall revenue growth, even while usage expands.
Ramp still saw spending growth among median companies and the top 10 percent of spenders. But the gap is large. Median firms spend between $12 and $15 per employee per month on AI, while the top 10 percent spend between $600 and $1,000.
Kharazian questioned whether mainstream businesses will ever resemble the very highest spenders. AI coding agents are in strong demand in software and finance, but he said it is less clear that a doctor’s office or car dealership would need AI spending anywhere near $10,000 per employee per month.
Price cuts and smarter controls are changing the market
The reported slowdown does not appear to be driven by falling interest in AI. Kharazian said usage volumes are increasing. The pressure is instead coming from cheaper access and better cost management.
OpenAI cut API prices in September by half for GPT-6 Sol and GPT-6 Luna. Anthropic also reduced pricing for Fable 5.1 and Opus 5.5. According to the source article, Anthropic said in a recent filing that its pricing decision reduced revenue growth by about $600 million. Both OpenAI and Anthropic have also offered enterprise discounts designed to secure customer commitments, though Kharazian said those discounts are affecting revenue targets.
He described the current market as favorable to buyers. Competition is not limited to the best-known model providers. Meta and Nvidia are introducing enterprise offerings with low-cost or open source models, and other open model companies in China and the United States are also competing for workloads.
Ramp’s data also points to Jev, a newer company building a lighter-weight tool for selecting among predetermined outputs. Kharazian said some AI use cases do not require a large, expensive model. In his example, a simple yes or no decision can often be handled by a cheaper system.
Companies are also learning how to govern consumption. Kharazian said more advanced AI users are setting defaults that prevent expensive models from being used for long-running tasks, placing budgets on employees, and shifting work from standard models to cheaper models through routers. In plain English, routers are systems that decide which AI model should handle a task based on cost, complexity, or performance needs.
Notion co-founder Akshay Kothari recently posted internal data that aligned with Ramp’s findings, summarizing the pattern as: “tokens go brrr. costs go ahhh.”
What this means for AI budgets and IPO expectations
For business leaders, the lesson is not to slow AI adoption automatically. The better reading is that AI operations are becoming a management discipline. Teams need people who can decide when a powerful model is justified, when a cheaper model is enough, and how AI work should fit into approved workflows.
That is a human-led task. Finance, IT, legal, security, and operational teams will need shared visibility into AI usage. Managers should expect more questions about who can spend, which tools are approved, what data can be used, and how productivity gains are measured.
The market implications are significant. Anthropic is preparing for an IPO, and Ramp’s data may complicate investor assumptions that leading AI model companies will keep expanding revenue at very high rates. Bloomberg reported Friday that Anthropic’s potential valuation had fallen from an estimated $2 trillion in early October to between $1.2 trillion and $1.3 trillion. The company is also reportedly expected to receive a high “junk” rating from credit agencies.
OpenAI has also lowered its proposed valuation to $1.25 trillion from $1.4 trillion and cut its latest fundraising goals by about $10 billion, according to the source article.
There was some improvement in October. Ramp saw token spending rise 8.9 percent from September 23 to October 23. Still, Kharazian said the forces limiting spend growth remain strong.
The open question is whether new categories of AI use will create another wave of demand. Coding agents became a major commercial use case over roughly the past year and a half, helping lift expectations for companies such as Anthropic and OpenAI. Some investors now expect personal agents to become the next major driver.
For employers, that is a possibility, not a plan. The immediate work is more concrete: train staff, design workflows, track usage, and make sure AI spending follows business value rather than novelty.
Reported by Hybrion Insights with reference to TechCrunch AI.
hybrion
Better together. Built for what's next.
Stay updated
Get insights on AI, teams, and the future of work.