Signal Inference cost fell 280-fold in two years while total AI spending rose faster still

Summary

Deloitte's infrastructure chapter opens with a paradox of AI economics. The cost of inference has fallen 280-fold over the past two years, yet enterprises are seeing overall AI spending grow explosively. The explanation is simple, because the volume of inference has climbed far faster than its unit price has declined. Tools built on large language model interfaces work for proofs of concept but become prohibitively expensive when rolled out across enterprise operations, and some companies now receive monthly AI bills in the tens of millions of dollars. Agentic AI is the largest contributor, since it runs inference almost continuously and can send token costs spiralling. The report argues this is forcing enterprises to recalculate where and how they run AI workloads at unprecedented speed.