Uber CTO says the company is moving past a “tokenmaxxing” approach to AI as efficiency becomes the focus

Uber is telling investors and the market that it is entering a more efficiency-focused phase of its artificial intelligence rollout. In comments reported by Yahoo Finance, Uber Chief Technology Officer Praveen Neppalli Naga said the company is coming to the end of what he called its “tokenmaxxing” era, a phrase used to describe a strategy centered on maximizing AI usage rather than optimizing outcomes.

Naga said Uber’s adoption of AI has accelerated quickly, describing it as having “quadrupled” since the company began scaling the technology more broadly. While the company has been expanding how it uses AI across its products and operations, he suggested that sheer scale is no longer the primary objective.

According to the report, Uber’s emphasis is shifting toward more efficient AI use, which implies tighter control over how much compute and model capacity is consumed per task, and how to ensure that AI contributes measurable improvements to products or costs. In other words, the goal becomes using AI in ways that deliver the greatest return rather than simply increasing volume of AI activity.

Uber has not, in the Yahoo Finance reporting, detailed what specific systems or teams are being targeted by the efficiency shift, nor has it provided metrics that would quantify how efficiency is being measured. The company also did not outline any new timelines, cost benchmarks, or performance targets in the excerpt referenced by the report.

The background for Uber’s decision is that AI adoption across the technology industry has rapidly moved from experimentation to operational deployment. As companies scale models to production, the economics of serving AI, including latency, infrastructure costs, and the amount of output or “tokens” generated per request, can become major drivers of overall spending. That dynamic is particularly relevant for consumer platforms and logistics networks that may process large numbers of real-time requests.

For Uber, the challenge is to translate broader AI adoption into business outcomes, such as improving matching and routing, improving customer and driver support, or increasing the quality of automated decisions. But the reported comments stop short of specifying where efficiency gains will show up first, or whether the company will reduce AI usage in some areas even as it maintains coverage in others.

As the company moves from scaling to optimization, the next question for investors will be how Uber defines “efficient” AI use in operational terms and whether it can connect efficiency efforts to financial impact. In the absence of disclosed targets in the reporting, watchers will likely look for additional detail in future earnings updates, product releases, or engineering-focused communications from Uber on AI governance and cost management.