Cost ↓50% × Utility ↑3× × Tokens/task ↑ = Volume ↑~6×Modeled mechanism
Indexed Q1 2024 = 100. Cost basket falls, agent ARR rises, inference volume rises fastest — multiplicative, mechanism estimate.
Cost basket
−30% YoY
Anthropic + OpenAI + Google composite, 2024–25
Agent ARR
~4× / 14mo
Cursor $500M→$2B; Copilot 77k orgs
Inference volume
~6× by Q1'26
Adoption × tokens-per-task ÷ cost — multiplicative
Observed: Q1'24–Q1'26 run-rateThe "6× — not 1.5×" reframe
Cost −30%, adoption ×4, reasoning models burn 3–10× more tokens per task. Multiplicative, not additive.