Uber Expands AWS Partnership with Custom Chips

Uber announced an expansion of its partnership with Amazon Web Services on Tuesday, deepening its integration with AWS's custom silicon to enhance real-time ride-matching and explore AI model training. The move involves deploying AWS's Graviton4 processors for Uber's core infrastructure and piloting the Trainium3 chip for machine learning workloads, marking a significant step in the company's multicloud strategy.

At the heart of the expansion is Uber's Trip Serving Zones system, which matches riders with drivers in milliseconds across more than 40 million daily trips in 72 countries. Officials at Uber stated that shifting this workload to the Graviton4, an ARM-based processor optimized for high-throughput and low-latency computing, will improve responsiveness during demand spikes, such as rush hours or events. The system processes vast amounts of data on driver locations, user preferences, and route optimization, where even brief delays can affect user experience.

Separately, Uber is testing the Trainium3 accelerator for training AI models on its historical dataset of over 13.5 billion trips and data from more than 200 million monthly active users. This initiative aims to refine...