Trend Compute strategy shifts from cloud-first to workload-matched hybrid placement

Summary

Deloitte's third trend examines how moving AI from proof of concept to production exposes infrastructure strategies that were never designed for it. Inference costs have fallen 280-fold over two years, yet total AI spending is surging because usage has grown far faster than prices have dropped, and some enterprises face monthly bills in the tens of millions of dollars. Continuous inference by agentic AI is the largest cost driver. The report stresses that the challenge also involves data sovereignty, latency, protection of intellectual property and resilience. Its answer is neither pulling everything back on premises nor adding more cloud, but matching each workload to the right platform, with cloud for elasticity, on-premises systems for steady production inference and the edge for immediacy. It suggests on-premises deployment may become cheaper once cloud costs exceed 60% to 70% of the cost of acquiring equivalent systems.