Why AI price cuts of 80% still leave your cloud bill rising: the infrastructure cost shift

Short summary

AI model token prices are dropping fast, but total AI infrastructure bills keep rising due to the Jevons paradox: cheaper tokens unlock 10× more usage. The real cost shift is from ~80% tokens in 2024 toward ~70% runtime, memory, observability, and GPU costs — infrastructure that follows traditional cloud pricing curves, not model price wars. Teams need standard FinOps discipline: schedule idle agent runtimes, tune observability retention, route requests to cheaper model tiers, and add anomaly detection for AI resources.

• •Token price cuts drive usage up 10×, netting higher bills (Jevons paradox)
• •AI spend shifting from ~80% tokens to ~70% runtime/memory/observability/GPU
• •Standard FinOps levers (scheduling, rightsizing, retention tuning) now apply to AI infrastructure

Generated with AI, which can make mistakes.