Token spending growth nearly flatlined over the summer, rising just 1.05% in August and a mere 0.22% in September, according to a report published on X by Global Market Observer. The two consecutive months of minimal expansion signal that the AI spending boom may be losing steam. The report warns that the AI growth narrative is increasingly vulnerable as cheaper tokens fail to generate enough new demand to compensate for collapsing prices.
The price drop has been dramatic. Anthropic's realized token price fell more than 51% in just one month, reaching $0.83 per million tokens—the lowest level ever recorded. OpenAI's comparable price stands at only $0.45. The concentration of spending has also intensified: a tiny 1% of companies generate 80% of all business spending on OpenAI and Anthropic combined. For Anthropic specifically, the top 1% of customers produced 46% of revenue in August, up from 25% a year earlier. Even the biggest spenders are slowing down: their outlays climbed 11.7% month-over-month to $9,568 per employee in September, but that figure still sits 3.1% below July's peak of $9,873. Meanwhile, paid AI adoption across Ramp's corporate network inched up only 0.62 percentage points to 55.9% in September.
The report finds that growth now depends increasingly on existing customers spending more rather than new customers adopting AI. According to the analysis, frontier-model token demand has expanded only half as fast as overall token demand, highlighting a disconnect between headline growth and the most advanced AI products. The slowdown challenges the assumption that cheaper tokens will automatically generate enough additional demand to offset falling prices.
The core issue is a mismatch between supply and demand dynamics. Prices are plummeting as AI providers compete and scale up capacity, but usage isn't accelerating fast enough to make up the difference. The concentration of spending among a small handful of heavy users suggests that AI has saturated its early-adopter market without breaking through to mainstream corporate buyers at scale. The stagnation in adoption rates reinforces this: when just over half of companies are paying for AI and that number barely budges month to month, revenue growth can only come from persuading existing users to spend more. The data suggests that strategy is running into limits, even among the most enthusiastic adopters whose spending peaked in July and hasn't recovered since.
The report's findings point to a fundamental test for the AI industry: whether falling costs can unlock mass-market demand or simply erode revenue without expanding the customer base. The combination of price collapse, spending concentration, and decelerating adoption suggests the sector may be entering a more difficult phase where hype gives way to harder questions about return on investment. The takeaway is blunt—AI providers can't count on cheaper tokens to drive growth if businesses aren't convinced the value justifies even minimal spending. This marks a critical juncture where the technology must prove its worth beyond the early believers, or risk a prolonged slowdown as pricing power evaporates and new customer acquisition stalls. The industry now faces a pivotal choice between doubling down on costly infrastructure in hopes of a delayed breakthrough, or confronting the possibility that current use cases may not justify the scale of investment markets have priced in.