Amazon’s Bedrock Shifts to Custom AI Chips, Reducing Reliance on Nvidia

Amazon Web Services has quietly reached a milestone in its AI infrastructure strategy. More than half of AWS Bedrock—its flagship cloud service for AI model access—now runs on custom Trainium chips rather than industry-standard Nvidia GPUs. The revelation came during an interview with AWS CMO Julia White, marking the first concrete disclosure about Amazon's in-house chip adoption.

The strategic shift carries significant financial implications. Trainium-based servers already undercut Nvidia-powered alternatives on price, potentially improving AWS margins while passing savings to cloud customers. Unlike general-purpose GPUs, these specialized processors accelerate both AI training and inference workloads—a design choice reflecting Amazon's vertically integrated approach to enterprise AI.