Tokens go brr, revenue goes flat: Inside the great AI spending stall

Companies are consuming more artificial intelligence than ever, but they are suddenly refusing to pay more for it. A report from financial platform Ramp shows that token spend flatlined over late summer even as token volumes rose, revealing a stark disconnect between usage and revenue. According to the data, token spend grew just 1.5 percent across August and September, a steep drop from a 42 percent jump in July.

This sudden deceleration suggests that price competition and savvier enterprise customers may be capping financial growth for foundational model giants just as Anthropic nears its public offering. Our read is that the market has transitioned from a land grab into a ruthless buyer’s market. Ramp’s lead economist Ara Kharazian notes that the slowdown is not driven by falling demand, but rather by an intense price war.

OpenAI and Anthropic have aggressively slashed API prices and offered enterprise discounts to lock in customers, moves that Anthropic filings show have cost hundreds of millions in revenue growth. Meanwhile, customers are getting thriftier. Enterprises are routing mundane tasks to ultra-cheap models, imposing internal budgets, and relying on lightweight routing tools like Jev to squeeze maximum efficiency out of minimal spending.

This thriftiness collides with a structural ceiling in the market.

Almost all heavy AI spending originates from venture-backed tech companies, with the top one percent of firms previously spending up to 10,000 dollars per employee monthly.

Yet everyday businesses like doctor offices or car dealerships have little use for such intense compute, meaning the hypergrowth seen among elite tech startups may not replicate across the broader economy.

Kharazian points out that while October saw a modest re-acceleration in token spend, strong structural headwinds remain as open-source models from Meta, China, and domestic competitors flood the market.

That reality check is already rippling through private markets.

Recent reports indicate that valuations for top AI labs have sunk significantly from earlier peaks, alongside lowered fundraising targets and looming credit scrutiny.

Investors are banking on personal agents or future breakthroughs to spark a new wave of enterprise spending, but skepticism remains high over whether these upcoming use cases can match the commercial productivity of coding agents.

For now, the era of unquestioned hypergrowth faces a hard ceiling, leaving model makers to figure out how to drive revenue when customers have learned how to use more intelligence while paying a fraction of the cost.