Amazon is exploring selling its custom Trainium AI chips to companies outside its cloud business, potentially creating a lower-cost competitor to Nvidia's dominant position—though both companies are growing fast enough that the market appears large enough for both.

What happened: Amazon has begun discussions with potential buyers about selling its Trainium accelerators (custom AI chips) for use in other companies' data centers, according to Bloomberg reporting. CEO Andy Jassy indicated in April there was "a good chance" Amazon would offer full racks of Trainium chips beyond its own cloud service within "the next couple of years." Amazon's broader custom-silicon business, which includes Trainium, Graviton, and Nitro chips, crossed a $20 billion（約3.2兆円） annual revenue run rate in the first quarter of 2026 and is growing at a triple-digit pace.

Why it matters: For years, Amazon has kept its chips internal, renting computing power through its cloud service rather than competing directly with Nvidia's GPUs in the open market. If Amazon begins external sales, it would offer customers a credible lower-cost alternative, which could put pressure on Nvidia's pricing power—the source of its exceptional profitability. Major AI customers like OpenAI, Anthropic, and Uber are already committed to significant Trainium capacity, suggesting real demand exists beyond AWS.

What to watch: No timeline or customer names have been announced yet, and Amazon's own actions suggest both companies can grow together: even as Amazon commits to selling Trainium externally, it is simultaneously planning to deploy more than one million Nvidia GPUs starting in 2026. Nvidia's data center revenue rose 92% year over year to a record $75.2 billion（約12兆円） in its fiscal first quarter of 2027, indicating the AI accelerator market is expanding fast enough to support multiple suppliers.