Levie contends that as AI token costs drop, consumption will increase for a wider range of tasks, leading to higher inference demand. This dynamic ensures that anything lowering token costs will boost GPU and infrastructure utilization, benefiting providers.
The Jevons paradox suggests that efficiency gains in AI will increase, not decrease, the total demand for compute infrastructure, maintaining high utilization for GPU providers.