Ramp finds AI spending growth slowed despite rising usage, before an October rebound

Businesses are using more AI without increasing their spending at anything like July’s pace. Ramp findings published by TechCrunch on November 6 show spending on AI tokens grew just 1.5% across August and September, before a partial October rebound. Ramp’s economist points to cheaper prices and more disciplined customers—not falling demand—as the explanation.

A rebound, but not a return to July

July had brought a 42% increase in token spending from the previous month. The subsequent 1.5% gain covered August and September together, making the contrast especially sharp: those figures describe different-length periods, not three equivalent monthly growth readings.

The latest window offered some relief. Spending rose 8.9% from September 23 to October 23. That was a renewed acceleration, although well short of July’s growth rate. Ara Kharazian, Ramp’s lead economist, told TechCrunch that the forces restraining spending remained strong despite the pickup.

The distribution of spending helps explain why. Kharazian said the most AI-intensive 1% of firms eased back in August and September, slipping below summer peaks of roughly $10,000 per employee per month. Spending among median firms and the top 10% grew, but not enough to offset that change among the heaviest users.

Customers stretch their AI budgets

Kharazian’s explanation separates the amount of AI being used from the bill attached to it. He attributed the divergence to model price cuts, enterprise discounts and customers learning to manage workloads more cheaply. On that reading, slower spending growth does not mean businesses have lost their appetite for AI.

He described a buyer’s market in which OpenAI and Anthropic face competition from inexpensive and open models. Enterprise discounts intended to secure customer spending also reduce what providers collect. Competition therefore works alongside customers’ own efforts to get the same work done for less.

More experienced AI customers have several ways to tighten that bill, Kharazian said:

• Set defaults that keep expensive models away from long-running tasks.
• Give employees spending budgets rather than leaving usage uncapped.
• Use routing tools to shift work from lower-cost models to even cheaper alternatives.

Not every task needs a sophisticated model, either. Kharazian pointed to simple yes-or-no decisions as work that can move to cheaper tools. His argument is that becoming more capable at using AI can make a company more economical, rather than automatically making it a bigger spender.

The next growth engine remains a bet

There is also a limit to assuming ordinary businesses will eventually spend like the most enthusiastic buyers. Kharazian contrasted demand for coding agents in software and finance with the needs of a doctor’s office or car dealership. He questioned whether those businesses would need anything approaching the top 1%’s spending per employee.

New products could still change the trajectory. Kharazian acknowledged the possibility of fresh uses that drive further spending, but was skeptical that personal agents would deliver the same economic value as coding agents. That leaves two competing forces in his outlook: new reasons to use AI, and increasingly effective ways to pay less for existing work.