Amazon custom silicon business reaches $20B annual run rate

Amazon announced its custom silicon business surpassed a $20 billion annual run rate with over 100% year-over-year growth. The division includes Graviton general-purpose processors, Trainium AI training chips, and Nitro security processors. Major multi-year commitments from OpenAI, Anthropic, Meta, and Uber drive the expansion.

Infrastructure economics shift

Amazon's custom chip business now ranks among the top three datacenter chip providers globally, reshaping infrastructure economics for AI labs. Previously, NVIDIA's dominance in GPU supply created a bottleneck for model training and inference. Custom silicon from Amazon, Google, and others reduces this dependency while lowering per-unit costs for large-scale deployments. The 100% YoY growth rate reflects both increased adoption and capacity expansion.

Supply chain implications

The shift toward custom silicon affects the entire AI supply chain. Training clusters using Trainium chips and inference systems using Graviton processors reduce reliance on NVIDIA's supply constraints. This diversification accelerates as frontier labs scale from billions to trillions of parameters, where custom silicon economics become critical to profitability.