Uber is strengthening its infrastructure and expanding its artificial intelligence capabilities on Amazon Web Services (AWS). The company already uses AWS Graviton instances to support its Trip Serving Zones, the real-time infrastructure that underpins each trip and delivery, and has begun running AI model training pilots on Trainium. This evolution of its technological infrastructure allows it to more quickly connect drivers and delivery people with demand, optimize global management and offer more intelligent and personalized experiences to millions of daily users.
Every time a user opens Uber and requests a service, a complex chain of decisions is triggered in milliseconds. Determining which driver is closest, what the best route is, or how long a trip will actually take requires a robust, flexible, and highly scalable infrastructure. This infrastructure is key to ensuring optimal performance even during peak hours or large events, where demand skyrockets.
Graviton powers real-time travel infrastructure
Trip Serving Zones are an essential part of Uber’s operational infrastructure, ensuring that each trip or delivery is frictionless. This system requires processing millions of predictions and location data in real time, relying on an infrastructure capable of operating in milliseconds.
To do this, Uber is expanding its use of AWS computing, storage and networking infrastructure. Migrating more workloads to AWS Graviton strengthens your infrastructure, allowing you to reduce energy consumption and scale quickly in the face of demand spikes. This infrastructure optimization not only decreases latency, but also improves cost efficiency. Thanks to Graviton’s high performance, Uber can perform real-time calculations within its infrastructure, accelerating the matching between passengers and drivers without compromising reliability, availability or security.
“Uber operates at a scale where milliseconds count,” said Kamran Zargahi, vice president of engineering at Uber. “Evolving our Trip Serving infrastructure on AWS gives us the flexibility to improve matching speed and seamlessly manage peak demand.”
Evolution of AI infrastructure with Trainium
Uber has also begun testing AWS Trainium as part of the evolution of its AI infrastructure. These models analyze data from billions of trips and deliveries, requiring high-performance infrastructure capable of supporting massive training loads.
Graviton4 and Trainium3 are chips custom designed by Amazon for AI computing and training
Training AI at this scale requires advanced computing infrastructure, and Trainium offers an efficient and cost-effective alternative to hardening that infrastructure. As models learn and evolve, this infrastructure enables faster connections, more accurate arrival estimates, and increasingly personalized recommendations to users around the world.
“By beginning to test our AI models in Trainium, we are laying the foundation for the technological infrastructure that will make every experience with Uber smarter, allowing us to focus on what matters most: the people who use Uber every day,” Zargahi added.
For his part, Rich Geraffo, vice president and general manager of AWS for North America, highlighted: “Uber is one of the most demanding real-time applications in the world, and we are proud to be part of the infrastructure that supports its global operations. Our infrastructure helps ensure the reliability that hundreds of millions of users rely on, as well as power the AI-based experiences that will define the future of transportation and on-demand deliveries.”
