AI Token Spending Stalls Amid Steep Price Declines
Ramp data shows corporate AI spending has stalled as price drops for frontier models offset rising token volumes, sparking warnings of a potential investment bubble.
Corporate spending on AI labs has stalled despite rising token volumes, according to data from Ramp tracking 70,000 U.S. businesses. The stagnation is driven by significant price declines for frontier models and increased efficiency, which have lowered the blended cost per million tokens.
Anthropic saw its realized token price drop from over $1.70 in mid-September to $0.83 by October, while OpenAI's index reached $0.45. This deflationary trend is compounded by the rise of cheap open-source alternatives. Spending is also highly concentrated, with 1% of companies accounting for 80% of all business expenditure. Anthropic is particularly vulnerable to this concentration, with two unnamed clients generating nearly a quarter of its 2025 revenue.
Economists from Apollo and Goldman Sachs warn that these trends may signal an AI bubble. They suggest that declining returns on invested capital could impact hyperscalers such as Amazon, Alphabet, Meta, Microsoft, and Oracle. The current data center construction models may be unsustainable if compute demand remains highly price-sensitive and fails to outpace the rapid decline in token costs.
