Uber Shifts Gears, Expands AI Workloads to Amazon's Custom Chips

Amazon Web Services secured a significant expansion of its cloud contract with Uber this week. The ride-sharing giant plans to run more of its core services on AWS, with a specific focus on Amazon's custom silicon. The deal involves broader adoption of the Graviton processor for general computing and a new pilot program testing Trainium3, Amazon's specialized chip for training artificial intelligence models.

This move is notable given Uber's recent cloud strategy. In 2023, the company announced large, multi-year agreements to migrate its infrastructure from its own data centers to Oracle Cloud Infrastructure and Google Cloud. As recently as December, Uber highlighted its use of Ampere's ARM-based chips within Oracle's cloud, calling the shift a "dual challenge."

The context makes Amazon's win more pointed. Oracle, once a major investor in Ampere, sold its stake late last year. Oracle founder Larry Ellison stated that in-house chip design was no longer a strategic priority, opting instead for large-scale purchases from suppliers like Nvidia. Amazon, however, is betting heavily on the opposite approach, designing its own processors to differentiate its cloud service.

By choosing to trial Trainium3, Uber joins a growing list of prominent tech firms, including Anthropic and Apple, that are increasing their AWS commitments specifically for access to Amazon's AI-optimized hardware. Amazon CEO Andy Jassy noted in December that the Trainium chip line has already grown into a multi-billion dollar business. For Uber, the decision appears to be a pragmatic shift toward the infrastructure it believes will best power its future services, even if it complicates the narrative of its earlier cloud partnerships.

Source: TechCrunch