AI
AI spending growth stalls despite rising token usage, threatening vendor revenue forecasts
New data from Ramp reveals that while companies are using more AI, they're paying less for it—a troubling sign for Anthropic and OpenAI as both prepare to go public.
TechCrunch — AI
Usage of AI tools continues to climb, but company spending on these services has largely flatlined. Fresh analysis from Ramp reveals a dramatic deceleration in spending growth that contradicts investor assumptions about the sector's trajectory, particularly as major AI firms approach their own public offerings.
Spending growth collapses while usage accelerates#
The gap between token consumption and spending tells the story. During August and September, token spend increased just 1.5%—a stark reversal from July's 42% month-over-month jump. Meanwhile, the actual volume of tokens being processed continues to climb, pointing to a troubling dynamic: companies are using more AI but getting it cheaper.
Ramp's chief economist captured the core problem: the highest-volume AI adopters—the top 1% of firms—have recently stumbled in their expenditures, falling just short of the $10,000-per-employee monthly threshold they hit over the summer. This slowdown among the deepest spenders is powerful enough to drag down overall revenue growth, even as token consumption continues rising.
The paradox: competition and savvy customers#
What's driving the spending slowdown isn't weak demand. It's a hypercompetitive market reshaping buyer dynamics. Both OpenAI and Anthropic have slashed prices substantially. OpenAI cut API rates in half for its GPT-6 models in September, forcing Anthropic to respond with steep discounts on its own offerings—a move the company disclosed will reduce annual revenue growth by roughly $600 million. Enterprise-wide discount deals intended to lock in long-term customers are instead simply reducing per-unit economics.
The competitive pressure extends beyond the two leaders. Meta and Nvidia are launching enterprise-grade offerings built on low-cost or open-source models. Lightweight alternatives like Jev—a tool for selecting among predefined outputs—are gaining traction as top performers among Ramp users, enabling even cheaper workloads.
Customers learning to optimize#
As adoption deepens, organizations are becoming more sophisticated about cost management. They're deploying routing tools to funnel expensive tasks to cheaper models, setting employee budgets, and relying on system defaults that avoid costly long-running processes. The result: meaningful work getting done on progressively smaller bills.

