Uber scales AI to 9.4x weekly requests with costs flat
Technology
Uber has massively boosted its AI game, handling 9.4 times more weekly requests since February 2026, but without spending extra.
Even as demand soared, Uber kept its AI costs flat, thanks to smarter tech moves.
Processing 1,000 requests is now nearly 34% cheaper from its April peak, and per-session costs have dropped by more than one-half since June.
Uber uses model routing and caching
AI now powers more than 70% of code changes at Uber and tackles more than 30,000 tasks every day.
Their secret? Matching each task with the right AI model: simple jobs go to budget-friendly models, while tougher ones get the heavy-duty treatment.
By capping session sizes and reusing data for longer (thanks to a smarter caching system), Uber scaled up fast without breaking the bank.