The cost of AI inference dropped about 280-fold in under two years — from roughly $20 per million tokens in late 2022 to around $0.07 by late 2024.
Source: Stanford HAI, AI Index Report 2025.
Why it matters: Collapsing inference costs — measured per million tokens — are a primary driver of enterprise adoption. Use cases that were uneconomical in 2022 are now trivially cheap, continually expanding what AI can profitably be applied to.
Drill down: see the full Generative AI Statistics 2026 and AI Infrastructure & Compute Statistics 2026 roundups. Related metrics: 23% scaling agentic AI · $252.3B corporate investment.