Amazon is preparing to sell its own Trainium AI accelerators to other companies, a move that would push its custom silicon beyond AWS and put more pressure on Nvidia’s grip on the AI hardware boom. The plan is still in talks, but the direction is clear: Amazon wants Trainium to become a product, not just an internal weapon.
That is a bigger deal than a simple chip shipment. Cloud giants are increasingly behaving like chip vendors because AI demand is pulling data-center buying power in two directions at once: toward the biggest hyperscalers and toward regional operators that want local infrastructure. Google has already taken a similar step with its Tensor processors, and Amazon is following the same playbook with its own custom accelerator stack.
Trainium sales move beyond AWS
Amazon currently uses Trainium in AWS data centers, where the chip has already been adopted by customers including OpenAI, Anthropic, and Uber. The company says Trainium 3, which began shipping earlier this year, is largely sold out, and interest is already building around Trainium 4, due next year. In other words, Amazon is trying to sell what it can barely keep on the shelf.
Peter DeSantis, who leads Amazon’s AI efforts, said the company has started discussions with data-center operators about supplying the chips, though he did not name any potential buyers. He argued that AI infrastructure is developing quickly and that Amazon wants to attract more customers, while brushing off any concern that external sales could hurt AWS.
Why Amazon thinks it can afford the gamble
The bet rests on a simple reality: AI compute remains scarce, and that shortage has created room for every serious supplier to try its luck. The boom has also fueled demand for sovereign cloud services in Europe and other regions that prefer local infrastructure, which gives Amazon another reason to package Trainium for third parties rather than leave it locked inside AWS.
Amazon is not starting from zero either. DeSantis said the company has added more Graviton processors to its systems over the past three years than any other chip type, and that it recently began supplying those general-purpose CPUs to Meta. That suggests Amazon is steadily building a broader hardware business, one that could help it defend margins even if Nvidia continues to dominate the premium AI chip market.
Amazon stock rises after Trainium sales news
Investors liked the story enough to push Amazon shares up 1.8% to $241.82 after the news. That reaction makes sense: selling Trainium externally could turn Amazon’s silicon investment into a new revenue stream, while also tightening its grip on AI customers who want alternatives to Nvidia without building their own chips from scratch.
- Trainium 3 is already largely sold out.
- Trainium 4 is expected to debut next year.
- Amazon says AWS has not been hurt by the rise in demand for sovereign and specialized AI clouds.
The open question is whether Amazon can turn this into a real second business or just another supply-constrained side hustle. If Trainium keeps gaining traction outside AWS, Nvidia gets a more credible challenger; if not, Amazon still ends up with a stronger chip stack inside its own cloud. Either way, the era of hyperscalers staying politely in their lane looks finished.