A surging global demand for localized artificial intelligence computing has triggered a strategic shift at Amazon. The company is actively preparing to sell its custom-designed Trainium AI chips directly to third-party companies for use in outside data centers. This decision marks a fundamental departure from Amazon’s historical approach of restricting its silicon strictly to its own Amazon Web Services ecosystem.
The push for sovereign AI infrastructure, particularly from European regulators and enterprises demanding local data processing, served as the macro-catalyst for the expansion. Amazon’s senior vice president of AI models, Peter DeSantis, confirmed the external sales strategy at the VivaTech conference in Paris in mid-June 2026. The move escalates an already complex hardware rivalry. Amazon’s custom-chip division currently operates at an estimated $20 billion annual revenue run rate.
Despite the direct competition, the companies remain deeply intertwined. As of mid-2026, Amazon is still one of Nvidia’s largest corporate clients. Amazon has committed to deploying 1 million Nvidia GPUs by the end of 2027 to satisfy the immediate compute requirements of its cloud customers.
The Graviton Playbook and the Cloud Hardware Shift
Alphabet initiated a similar strategy in April 2026 when it announced it would deliver its rival Tensor Processing Units to a select group of external customers. Amazon’s hardware expansion directly mirrors its highly successful 2018 infrastructure strategy.
During that period, the company steadily eroded Intel’s server processing monopoly by introducing its custom, highly efficient Graviton CPUs to the market. By offering Trainium chips directly to external data centers, Amazon aims to replicate that disruption in the AI accelerator sector.