The price of artificial intelligence tokens has fallen sharply in the United States, raising fresh questions over whether the booming demand expected to support the industry’s vast infrastructure spending will materialise.
Businesses are now paying an average of 68 cents per million tokens, down from a peak of $1.15 in March, according to data from corporate spending platform Ramp. The 41 per cent decline has come despite the release of increasingly capable models by leading AI companies.
Ara Kharazian, Ramp’s chief economist, said the figures represented a “crack in the AI thesis” rather than evidence that the market was collapsing. But he warned that tokens were beginning to behave more like a commodity, with customers switching between increasingly similar and cheaper products.
AI token prices fall as companies trade down
The share of usage going to frontier models fell from about 53 per cent in early August to 45 per cent by September, Ramp said. Among the largest corporate users, which account for roughly 80 per cent of OpenAI and Anthropic’s enterprise revenue, spending per employee dropped by nearly 10 per cent in August.
The most AI-intensive 1 per cent of companies still spend about $7,200 per employee each month. That is well below Nvidia chief executive Jensen Huang’s suggested target of $250,000 a year for a highly paid engineer, and the figure is now falling rather than accelerating.
Kharazian said the decline reflected both price cuts by AI laboratories and a shift by customers towards less expensive models. OpenAI cut the price of its GPT-5.6 Luna model by 80 per cent in July, while Anthropic has also reduced prices. OpenAI said Luna now costs 20 cents per million input tokens and $1.20 per million output tokens. ([openai.com](https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/?utm_source=openai))
OpenAI finance chief Sarah Friar said at a Goldman Sachs conference in San Francisco on Monday that the price reduction had been followed by roughly a tenfold increase in usage. She also said the company wanted to move enterprise customers away from paying for individual tokens and towards charges based on completed work. ([finance.yahoo.com](https://finance.yahoo.com/technology/ai/articles/openai-cfo-sarah-friar-says-123936035.html?utm_source=openai))
That dynamic is intensifying competition between the major providers. Since August 1, OpenAI’s effective price has fallen 38 per cent to 48 cents, while Anthropic’s has fallen 22 per cent to 90 cents, according to Ramp. Anthropic has generally maintained a higher price, suggesting it retains some pricing power, although Kharazian said that advantage was narrowing.
The pressure is not limited to American companies. Only 3.6 per cent of businesses tracked by Ramp use open-source or Chinese models, but intense competition from providers including DeepSeek, Tencent and Alibaba is also contributing to lower prices across the market.
The trend has unsettled the assumptions behind the huge investment in data centres and specialist computing providers. Morgan Stanley has warned that as much as $300 billion in bonds used to finance so-called neocloud companies could be vulnerable if token prices fail to keep pace with the cost of building and operating infrastructure.
Citadel Securities has separately pointed to a widening divide between frontier AI, used mainly by a small number of wealthy technology companies, and the cheaper, everyday systems being adopted elsewhere. The concern for investors is that falling prices may increase usage but reduce the revenue available to repay the debt accumulated during the industry’s expansion.
“Prices are relative to the other products available on the market,” Kharazian said, arguing that businesses were still willing to pay for powerful systems when the benefit justified the cost. But with mid-tier models becoming more capable, companies are increasingly setting defaults that steer staff away from the most expensive tools.
For OpenAI and Anthropic, the immediate challenge is to turn lower prices into enough additional usage to preserve growth. The figures suggest demand for AI remains substantial, but that customers are becoming more selective over how much they pay for it.
