route: compounding
per-token serving cost falls fastslow | pricing power
Gross-margin waterfall: inference unit economics reroutes run-rate to profit or burn gross-margin waterfall — the unit-economics gate inside the run-rate revenue run-rate tokens served compute cost per inference cut margin outlook — growth ≠ profit serving cost falls fast → slow pricing power holds → match cuts unit-economics lever scale cuts cost price war → margin durable compounding reported Q2 2026 profit operating leverage compounds burn treadmill margin compresses run-rate compounds losses single gate: serving-cost trajectory vs pricing power reroutes the stream branches are scenarios — the reported first profit is not company-confirmed