Uber's chief financial officer, Balaji Krishnamurthy, stated that the company maintained stable AI spending in the second quarter by utilizing lower-cost models, optimizing defaults for various use cases, and enabling employees to better manage their AI expenses. This approach has led to a decrease in cost per token.
Uber's approach to managing AI costs demonstrates a practical strategy for integrating revolutionary technology into existing business operations without incurring unsustainable expenses, potentially serving as a model for other companies navigating similar challenges.
Uber is strategically managing its artificial intelligence spending by implementing a multi-faceted approach that includes utilizing more cost-effective AI models and empowering employees to monitor and control their AI expenses. CFO Balaji Krishnamurthy reported that these measures have successfully stabilized overall AI expenditure during the second quarter, even as the company's adoption of AI technologies continues to grow.
Krishnamurthy explained that by setting optimized defaults for various use cases and enhancing employee awareness of AI costs, Uber has seen a reduction in its cost per token over recent months. This contrasts with an earlier incident where CTO Praveen Neppalli Naga's comment about exhausting the 2026 Claude Code budget within months went viral, highlighting potential cost concerns.
CEO Dara Khosrowshahi emphasized that Uber's AI strategy focuses on incremental improvements across its ecosystem rather than seeking a single revolutionary breakthrough. He cited destination suggestions within the app, which accurately predict user destinations three-quarters of the time, as an example of these numerous small enhancements. Khosrowshahi believes these cumulative improvements will drive future growth.
The company is also leveraging AI to streamline internal operations, including a recent layoff of approximately 10% of its customer service staff, attributed to efficiency gains from AI. Krishnamurthy highlighted customer service as an area where AI can improve support quality while reducing costs. Uber is also exploring AI applications in finance and human resources through "agentic pods."
Consumer-facing AI applications are also progressing, though not expected to be deployed all at once. While Uber facilitates hundreds of thousands of rides weekly in self-driving cars, this represents less than 0.5% of its total weekly trips. Khosrowshahi contrasted this with the widespread adoption of AI in online search, noting that the physical application of AI in robotaxis requires extensive real-world testing and regulatory approval.
Uber is also experimenting with features like Cart Assistant, which helps users create grocery shopping carts from recipes or lists. Users of this feature have demonstrated cart sizes twice as large as those who do not use the AI tool. Khosrowshahi anticipates AI will contribute to increasing average order sizes and enhancing service quality and reliability.