Amazon's AWS may sell Trainium chips to challenge Nvidia

Amazon's AWS is in talks to sell its Trainium AI chips to other companies, AWS AI chief Peter DeSantis told Bloomberg, though DeSantis declined to name potential buyers. The company also told TechCrunch those conversations are in early stages and trace back to comments from CEO Andy Jassy in his shareholder letter in early April.

In that letter Jassy wrote, "If our chips business was a standalone business, and sold chips produced this year to AWS and other third parties (as other leading chips companies do), our annual run rate would be ~$50 billion." He added that demand has been so strong the company has considered selling racks of chips to third parties in the future.

What has AWS actually said about selling Trainium?

AWS says it has historically declined requests to sell its AI chips directly, but company leaders now describe third-party sales as possible; the talks are described as being in the early stages. Peter DeSantis confirmed AWS is talking with outside parties, and an AWS spokesperson reiterated that while the company has historically declined those requests, "it’s quite possible we’ll sell racks of them to third parties in the future."

The idea grew out of Jassy’s April shareholder letter, which framed Trainium demand as intense: Jassy said current Trainium capacity had "sold out almost instantly," and that capacity for the next chip, Trainium4, had also sold out and "won’t even be available for more than a year."

How big a challenge to Nvidia would this be?

If AWS sold chips as a standalone business, Jassy estimated it could produce an annual run rate of about $50 billion, while Nvidia is described in the piece as "currently on a $326 billion revenue run rate." That scale would not topple Nvidia, the article notes, but it would place Amazon’s chip ambitions closer to the size of Intel’s annual revenues and make AWS a material competitor in parts of the AI chip market.

The story contrasts Jassy’s $50 billion figure with Nvidia CEO Jensen Huang’s own framing that he has "found a brand new $200 billion market for Nvidia in selling CPUs for AI, not just GPUs," underscoring that both companies see new revenue opportunities tied to AI compute. The TechCrunch piece points out a structural constraint: AWS could only sell chips externally if it can both meet internal demand and secure additional manufacturing capacity from partners such as TSMC, which recently supplanted Apple as TSMC’s largest customer.

Why it matters

AWS moving from using Trainium internally to selling it externally would shift its role in the supply chain: from a cloud service provider that monetizes chips indirectly via compute, storage and related services, to a direct hardware competitor in data-center silicon. That would change where value accrues: today AWS captures chip-driven revenue across tokens and a host of ancillary services; selling hardware would monetize the silicon itself and expose AWS to foundry constraints and direct competition with established chip vendors.

Shifting to hardware sales would also pressure the foundry layer. TechCrunch highlights that AWS would likely need to expand production through partners such as TSMC, and doing so while TSMC’s largest customer position recently shifted could complicate AWS’s ability to scale quickly.

What to watch

Watch for any formal announcements of rack or chip sales from AWS and whether AWS publicly names customers or partners; that would move the talks from "early stages" into commercial commitments. Also watch Trainium4 availability: Jassy said Trainium4 "won’t even be available for more than a year," so manufacturing capacity and delivery timelines for that chip will be a key signal of whether AWS can both serve internal demand and sell externally.