Particle.news
Download on the App Store

Uber Expands Real-Time Workloads on AWS Graviton4, Pilots AI Training on Trainium3

The shift signals a bid for faster service with lower compute costs using Amazon-designed chips.

Overview

  • Uber is moving more Trip Serving Zone systems onto AWS Graviton4 and has begun piloting AI model training on Trainium3, extending its cloud partnership with Amazon Web Services.
  • Trip Serving Zones run the split-second decisions behind rides and deliveries, including driver matching, routing, and time estimates for millions of requests each day.
  • Uber and AWS say Graviton4 helps cut latency, handle demand spikes without disruption, and reduce energy use and cost, though these are vendor-reported results.
  • The Trainium3 pilot targets training models for faster matches, more accurate arrival times, and more tailored in-app suggestions, with outcomes still in testing.
  • Graviton4 is an Amazon-built CPU for general compute and Trainium3 is its AI training chip, a pairing AWS is promoting as a custom alternative to standard processors as it seeks more enterprise AI work.