Amazon weighs selling its Trainium AI chips to other companies in bid to widen use beyond its own data centers

Amazon is exploring the possibility of selling its Trainium AI chips directly to other companies, a step that would push the Seattle retailer’s custom hardware further into the wider artificial intelligence infrastructure market, according to a report from Bloomberg News carried by Yahoo Finance.

The report, citing an interview with Amazon’s AI chief, Peter, says Amazon is in talks to offer Trainium for use in other firms’ data centers. The stated idea is not simply internal deployment, but commercialization of the chips for outside customers, which Amazon has so far been able to provide mainly through its own cloud services.

Trainium is Amazon’s custom-designed chip aimed at accelerating the training of machine learning models, a core workload for generative AI systems. Training chips are typically used to run large-scale learning jobs, often in massive clusters, where software and hardware are tuned together to improve performance and reduce total cost of running AI workloads.

For Nvidia, which is widely associated with AI data-center accelerators, the prospect of a major cloud provider distributing its own training silicon to external customers would represent a fresh competitive pressure point. Even if Trainium’s reach grows gradually, the mere fact that a large buyer of AI compute could become a chip supplier changes the bargaining dynamics around capacity and supply planning for developers and enterprises building AI systems.

The report characterizes the discussions as early-stage talks rather than a finalized product announcement. It does not, in the account cited by Yahoo Finance, provide details on customer eligibility, pricing, contractual terms, expected timelines, or the exact integration support that Amazon would offer to buyers outside its cloud environment.

From a market perspective, the AI chip supply chain has increasingly been shaped by vertically integrated players, where cloud platforms both buy accelerators and develop custom alternatives to optimize costs. If Amazon were able to convert Trainium into a product line usable beyond its own infrastructure, it could announcement confidence in not only the chip design but also the broader ecosystem required for external deployment, including orchestration software, cluster engineering, and ongoing hardware support.

Still, key questions remain unanswered in the cited report. Amazon did not disclose in the Yahoo Finance/Bloomberg account whether Trainium would be sold as standalone hardware only, bundled with managed software and support, or offered alongside other chips in Amazon’s stack. It also did not specify which types of customers Amazon has targeted first, such as enterprises building on-premise clusters versus firms already relying on cloud procurement.

Investors and industry watchers will likely look next for any concrete commercial details, including whether Amazon plans a pilot program, the scope of workloads it intends to support, and how it plans to position Trainium against incumbent data-center accelerators. The competitive impact will depend less on headlines and more on actual customer adoption, performance benchmarks in real deployments, and the commercial terms Amazon is willing to offer.