Amazon weighs selling its AI chips beyond AWS, a potential new front against Nvidia

Amazon Web Services is exploring ways to sell the AI chips it designs to other data centers, an approach that could put more pressure on Nvidia and other suppliers that dominate the market for AI accelerator hardware. In comments reported by Yahoo Finance, AWS CEO Andy Jassy framed the move as a significant business opportunity for Amazon.

The concept centers on making chips built for AWS workloads available to outside customers, rather than keeping them limited to Amazon’s own cloud infrastructure. If pursued, the strategy would represent a more direct challenge to Nvidia’s role supplying the core computing accelerators used to train and run AI models at scale.

The reported talks are not described as a finalized commercial launch. Instead, they suggest AWS is evaluating whether it can distribute its own AI silicon more broadly through partnerships or sales arrangements, including to other data center operators that want AI acceleration without relying solely on third-party chip makers.

Jassy’s estimate, as reported, is that the opportunity could reach $50 billion for Amazon. That figure is presented as the potential value of selling AWS-designed chips externally, rather than as confirmed revenue guidance or a near-term order book.

For Nvidia, the threat would not be hypothetical. Its current position relies heavily on the demand generated by data center AI buildouts. Any move by AWS to increase external availability of its own accelerators could alter procurement patterns, even if Amazon still sources some components and software from an ecosystem tied to Nvidia’s platform.

The wider sector context is that cloud providers have increasingly pushed to differentiate their offerings with proprietary hardware and software stacks optimized for AI workloads. AWS, Google, Microsoft, and others have invested in custom chips and infrastructure, aiming to improve cost efficiency and performance for their own services, while also controlling more of the supply chain.

Still, what Amazon would actually disclose about performance, pricing, and timelines remains unclear from the reporting. The Yahoo Finance piece, as characterized in the available material, does not provide details on which specific chip(s) would be offered, what technical specifications would be published, what customers would be targeted first, or when any products would be broadly available.

What to watch next is whether Amazon moves from exploratory discussions to a concrete product plan, such as naming chip families, stating support for key software and developer frameworks, and describing commercial terms for external buyers. Another sign will be how Nvidia responds, either by adjusting pricing, expanding partnerships with cloud customers, or emphasizing the value of its hardware and full-stack software for training and inference.