AI Token Prices Fall Below $1, Squeezing Frontier Labs

Large-language-model competition is shifting from capability gains toward cost and scale. Inference services are typically priced by token usage, and cheaper rates can accelerate enterprise adoption and broaden AI applications. But they also make it harder for frontier laboratories such as OpenAI and Anthropic to recoup spending on model training, chips and data centers, raising questions about the sustainability of their pricing power.

As of Sept. 2, 2026, benchmark pricing fell below $1 per million tokens for the first time, according to the latest report. The record low comes while fixed computing expenses, including GPUs, electricity and data-center capacity, remain elevated. That mismatch means rising usage may not translate proportionally into profit, putting added pressure on revenue growth and margins across the foundation-model industry.