Amazon CEO Andy Jassy confirmed that the company will continue to rely on Nvidia hardware for its AWS cloud platform. In an April 2026 letter to shareholders, Jassy noted that most AI development currently utilizes Nvidia chips, and AWS remains a primary destination for these workloads.
In August 2026, Amazon committed to purchasing an additional 2 million Nvidia GPUs for deployment between 2027 and 2028. This order adds to a previous agreement for 1 million units, bringing the total expected pipeline to approximately 3 million Nvidia processors.
The company's internal AI chip business, which includes the Graviton, Trainium, and Nitro lines, has reached an annual revenue run rate exceeding $20 billion. Amazon estimates that this figure could rise to $50 billion if the company decides to commercialize these chips externally.
Amazon anticipates that its Trainium line will generate savings of tens of billions of dollars in annual capital expenditures. The company aims to use these custom chips for routine inference and fine-tuning tasks, while relying on Nvidia for high-end model training and frontier inference.
The upcoming Trainium3 chips are expected to offer a 30–40% performance improvement. Amazon joins other major hyperscalers, including Google, Microsoft, and Meta, in developing proprietary AI accelerators to improve internal economics while maintaining partnerships with Nvidia.
