Uber's coo says ai token spending is getting harder to justify

Uber's COO Andrew MacDonald says the company is finding it harder to justify its AI token spending as costs rise without proportional returns [Business Insider]. The pushback comes as Uber scales AI across customer support, routing, and fraud detection—areas where token use has spiked but financial benefits remain murky.

MacDonald did not provide exact figures, but confirmed that some AI workloads now cost millions annually with unclear efficiency gains. He emphasized that teams must now defend AI budgets like any other operational expense, signaling a shift from the earlier 'move fast' AI adoption phase.

This isn't just an Uber problem. Companies like Meta and Salesforce have also tightened AI spending oversight in 2026, according to internal memos and earnings calls [Business Insider]. As models require longer context windows and higher throughput, token counts—and bills—balloon, especially in real-time systems processing thousands of queries per second.

The broader trend suggests AI is entering a cost-optimization cycle. Firms are now auditing model usage, replacing expensive LLMs with fine-tuned smaller models where possible, and enforcing token budgets per team. At Uber, one engineering group cut token use by 40% by switching from GPT-4 to a distilled 7B-parameter model for internal summarization tasks.

If even a well-resourced company like Uber is hitting ROI walls, it’s a warning sign for the wider AI infrastructure stack. The era of unchecked token consumption may be ending.