Key takeaways
- AI token spending growth plummeted to 1.5% in August-September from 42% in July, signaling that the acceleration driving startup valuations may be slowing.
- Price cuts from OpenAI and Anthropic, costing Anthropic $600 million in revenue growth, have created a buyer's market where increased usage does not translate to higher vendor revenue.
- Enterprises are optimizing AI spending through efficiency measures including routing cheaper models and employee spending caps, while competition from Meta, Nvidia, and open-source vendors limits pricing power.
- Anthropic's valuation fell from $2 trillion to $1.2-1.3 trillion and OpenAI cut its target from $1.4 trillion to $1.25 trillion, reflecting investor concerns about sustainable growth.
According to fresh data from Ramp, the acceleration in AI spending that defined 2024 and early 2025 has sharply decelerated. While organizations continue to deploy AI tools at scale, the revenue growth that powers the industry’s largest players is not keeping pace with usage growth, a divergence that has immediate implications for valuations and investor expectations as major AI companies approach public markets.
Token Growth Collapsed After Summer
The dramatic drop in spending momentum
Ramp’s spending data tracks the spending patterns of venture-backed technology companies and reveals the pace of AI adoption at the enterprise level. In July, token spending grew 42% compared to the previous month, but that trajectory reversed sharply. From August through September, token spend grew only 1.5%, marking a dramatic shift in momentum.
What the top 1% reveals about market saturation
The most aggressive AI adopters, what Ramp’s lead economist Ara Kharazian calls the top 1% of “AI-pilled” firms, showed signs of pulling back their spending during this same period. These firms had been spending nearly $10,000 per employee per month at their summer peak, but that figure slipped just below that threshold in August and September. For context, the median firm spends between $12 and $15 per employee per month, while the top 10% of companies spend between $600 and $1,000.
October data did show recovery, with token spending rising 8.9% from September 23 to October 23, but Kharazian cautioned that headwinds on spending growth remain persistent.
The limits of AI adoption across industries
Kharazian raised a fundamental question about the long-term trajectory: whether most organizations will ever need to reach the spending levels of the top 1%. AI coding agents have proven commercially valuable in software and finance sectors, he noted, but it remains unclear whether a medical practice or car dealership would justify similar per-employee spending. That uncertainty undermines the bull case for uninterrupted growth.
Price Wars Are Reshaping the Market
The slowdown in spending growth does not stem from weak demand for AI tools. Token volumes continue to climb. Instead, the industry’s largest vendors are engaged in aggressive pricing competition that is constraining revenue even as usage accelerates.
In September, OpenAI cut its API prices in half for GPT-6 Sol and GPT-6 Luna. Anthropic responded with significant price reductions for its Fable 5.1 and Opus 5.5 models. The company disclosed in a recent filing that these pricing cuts will reduce revenue growth by approximately $600 million.
Both companies have layered enterprise discounts on top of published price cuts, ostensibly to lock in long-term customer commitments. However, Kharazian observed that these discounts are achieving the opposite of their stated goal, eroding revenue targets rather than securing them.
A Buyer’s Market Takes Shape
The competitive landscape has shifted decisively in favor of customers. This dynamic mirrors historical cycles in Big Tech markets: ride-hailing companies like Uber and Lyft once offered deeply subsidized fares to build market share, then raised prices once competitive threats diminished. That strategy cannot work in AI, Kharazian argued, because the competitive moat never solidifies.
Meta and Nvidia have launched enterprise offerings built on efficient, lower-cost models or open source alternatives. Meanwhile, a growing roster of competitors, including open model companies from China and the United States plus newer entrants like Jev, a tool for routing requests between predetermined model outputs, now rank among top performers in Ramp’s data. For straightforward classification tasks that require only a yes-or-no decision, routing requests through cheaper models via tools like Jev yields further cost reductions.
Customer Sophistication Is Driving Efficiency Gains
Beyond vendor price cuts, enterprises are learning to operate AI infrastructure more efficiently. Kharazian observed that advanced users now employ strategies that were uncommon a year ago: setting spending caps on individual employees, defaulting to cheaper model variants for routine tasks, and using routing layers to steer high-cost workloads to lighter, less expensive models.
Akshay Kothari, co-founder of Notion, published internal spending data that aligned with Ramp’s findings. His characterization captured the dynamic plainly: “tokens go brrr. costs go ahhh.”
The implication is stark: more AI, deployed more broadly, generated at lower cost. That arithmetic benefits end customers but pressures the vendors whose revenue models depend on token volume times price.
Valuations Reflect Spending Reality
The Ramp data may explain recent revaluations at both Anthropic and OpenAI. According to Bloomberg reporting, Anthropic’s potential valuation fell from $2 trillion in early October to $1.2 to $1.3 trillion. Credit agencies are reportedly preparing to assign the company a high “junk” rating. OpenAI lowered its valuation to $1.25 trillion from $1.4 trillion and cut its latest fundraising targets by approximately $10 billion.
The Search for the Next Growth Driver
The bull case for AI spending has always rested on the emergence of new, economically productive use cases that would sustain exponential growth. AI coding tools emerged roughly 18 months ago and became the first major commercial application that justified billion-dollar valuations for their creators.
Investors are now speculating that personal agents will occupy that role next. OpenAI, SpaceXAI, and Meta have all released personal agent products, alongside smaller startups including Instinct. Kharazian expressed skepticism about whether personal agents, which lack the economic productivity of coding tools, could generate the sustained spending acceleration the industry requires.
“The projection is that new products will drive growth,” Kharazian said. “But based on what we can observe today and the products available commercially, AI spend has started to slow down.”
Source: TechCrunch
Frequently Asked Questions
Why did AI spending growth slow so dramatically?
Token spending grew only 1.5% in August-September compared to 42% in July. The slowdown stems from aggressive price competition between OpenAI and Anthropic, enterprise optimization tactics that reduce costs, and questions about whether most industries will justify large-scale AI spending.
How much revenue are AI companies losing to price cuts?
Anthropic disclosed that its pricing reductions for Fable 5.1 and Opus 5.5 will reduce revenue growth by approximately $600 million. In September, OpenAI cut its API prices in half for GPT-6 Sol and GPT-6 Luna.
What could reignite AI spending growth?
Investors speculate that personal agents could be the next major use case. OpenAI, Meta, and SpaceXAI have all released agent products. However, analysts remain skeptical that personal agents will have the economic productivity needed to sustain the hypergrowth the industry has experienced.
