Uber expands AWS contract and bets on Amazon chips, bypassing Google and Oracle

Uber expands AWS contract and bets on Amazon chips, bypassing Google and Oracle

Uber is expanding its contract with Amazon Web Services and moving more of its service functions to Amazon chips—a major victory for AWS in competition with Google Cloud and Oracle, which also vied for this business. The decision by the world's largest taxi aggregator signals to the market: Amazon's custom AI processors are beginning to seriously rival NVIDIA's standard GPU solutions in the enterprise segment. Amazon has been developing its own AI chips for several years.

Trainium is optimized for training neural networks, Inferentia—for their inference in production. Uber is switching part of its AI workloads to precisely these processors. Previously, these tasks required standard GPU clusters or computational resources from other cloud providers.

AWS positions its silicon as a cheaper alternative for inference tasks: with comparable throughput, the cost of inference on Inferentia is lower than on NVIDIA H100, which is critical for high-frequency production services. The contract expansion is aimed at real-time AI workloads: driver matching algorithms, demand forecasting by district, dynamic pricing, fraud detection systems, and anti-abuse filters. All these components require continuous inference of ML models with minimal latency.

Uber processes such workloads around the clock in dozens of cities simultaneously, and the cost of inference is one of the key line items in the company's cloud budget. Reducing this line by 20-30% means tens of millions of dollars in annual savings. For Google and Oracle, the news sounds like a public rebuke.

Both companies are aggressively expanding their AI infrastructure, poaching corporate clients. Google Cloud is betting on fourth-generation TPU and A3 clusters based on NVIDIA H100, Oracle—on supercomputers with A100/H100, positioning them as the most performant AI infrastructure in the cloud. Uber's choice in favor of AWS with its custom chips is a signal that the economics of AI infrastructure are beginning to work in Amazon's favor.

Amazon Web Services remains the largest cloud provider with approximately 30% market share by revenue. But AI has become a new field of competition: corporate clients are rebuilding their infrastructure and looking for ways to reduce inference costs. Each major transition to AWS custom silicon is a market signal and a marketing case for negotiations with other enterprises.

Uber, in this sense, is an ideal demonstration story for Amazon's sales team. Uber is among the most technically demanding platforms in the world. The company processes millions of rides daily in 70+ countries, manages Uber Eats and Uber Freight, and conducts large-scale R&D programs in autonomous driving.

This scale means that any platform choice is made after detailed technical and financial analysis. When Uber chooses a certain type of AI chip, it is not a marketing partnership—it is an engineering decision with multi-year consequences. The expansion of partnership with AWS is further evidence that the race for AI infrastructure has long gone beyond the familiar NVIDIA-versus-everyone-else confrontation.

Amazon, Google, and Microsoft are building their own processors, and major technology companies are forced to make a specific choice: whose silicon will form the basis of their next generation of AI. Uber chose Amazon.