Uber Partners with Amazon for AI Efforts, Leverages Custom Chips

Uber and Amazon have formed a partnership to enhance the company's AI efforts, utilizing Amazon's AI processing chips and cloud services. The collaboration aims to improve ride-sharing features and efficiency, with implications for the tech industry and Uber's competitors.

UBER announced on April 7 an expanded partnership with AMZN, moving core infrastructure workloads to AWS and beginning a pilot to train AI models on Amazon's Trainium3 custom silicon. The strategic significance extends beyond a typical cloud deal: Trainium3 runs at roughly 30 to 50 percent of the cost of comparable Nvidia H100 hardware, meaning Uber gains material compute cost advantages as it scales AI training across billions of rides and deliveries per year.

AWS Graviton4 processors will handle Trip Serving Zones — the real-time routing, dispatch, and matching logic that powers every ride request across Uber's global network. Moving this workload to Amazon custom silicon signals Uber's confidence in AWS's chip roadmap and deepens its infrastructure dependency at a time when hyperscalers are competing aggressively for enterprise AI workloads. Uber joins Anthropic, OpenAI, and Apple as companies that have expanded AWS commitments specifically because of Trainium.

For AMZN, Uber's adoption validates Trainium3's production readiness at extreme operational scale. Uber's data complexity — real-time AI inference for millions of simultaneous trips — makes it arguably the most consequential enterprise proof point yet for Amazon's custom chip ambitions, directly challenging Nvidia's dominance in AI training and positioning AWS as a credible alternative to GPU-based infrastructure.