Uber’s AI Spending Crisis: The End of the Tokenmaxxing Era and the Rise of Efficiency

Uber announced that it burned through its 2026 AI budget within months, signalling the end of the tokenmaxxing era with a fresh cost‑efficiency playbook.

AI Spending Collapse and Reassessment

Uber CTO Praveen Neppalli Naga disclosed that while employees were encouraged to use frontier tools like Anthropic’s Claude Code, the rapid budget depletion forced a "back to the drawing board" approach. By quadrupling the number of AI users, the company succeeded in lowering the cost per token.

Drivers Behind the Cost Decline

■Prompt caching was enhanced, eliminating redundant query processing.

■Default model settings were switched to more efficient models.

■Engineers gained real‑time visibility into hourly AI usage and costs.

■Token price fell by 90%, yet consumption rose 450% (Bain & Co. study).

Broader Industry Takeaways

■Magnificent Seven firms saw profit margins grow from 15% to 25%, while the rest of the S&P 500 recorded only a 10% increase.

■Treating AI efficiency as an engineering problem rather than a budgeting issue can reverse cost trends.

■Companies are shifting from a quantity‑first to a quality‑first token‑spending mindset.

Jevons Paradox and Future Risks

The 19th‑century observation by William Stanley Jevons that cheaper coal led to higher consumption now mirrors AI token economics: lower unit costs drive higher total spend. Apollo chief economist Torsten Slok notes that firms deploy more AI agents and automation as tokens become cheaper.

Uber’s revamped strategy does more than rein in AI expenses—it embeds a efficiency‑first engineering culture across the organization. In aviation and logistics, a similar shift could unlock substantial cost savings and service improvements, provided AI usage is tightly managed for both cost and performance. – Gökberk Uçar, Aviation Logistics and Cargo Specialist