Super-Linear Growth in Inference Demand (2023-2028)

The data reveals a striking pattern. Between 2023 and 2028, the projected inference demand index grows from 500 to 51,000, that more than a 100-fold increase in just five years. This growth is driven by two compounding factors: autonomous AI agents that execute longer, multi-step workflows, and the increasing use of test-time compute that generates more tokens per step. This super-linear growth pattern is what makes the inference bottleneck so critical. Unlike training, which is a one-time cost, inference costs scale with each user interaction, query, and autonomous agent action. As these numbers grow exponentially, the economic implications become staggering.