Amazon Looks Beyond AWS as It Prepares to Sell AI Chips Directly
by Michael Hicklen - 13 hours ago - 5 min read
Amazon is preparing to take its AI chip ambitions beyond its own cloud.
The company is in talks to sell its custom Trainium AI chips to outside customers for use in their own data centers, a move that would put Amazon in more direct competition with Nvidia, AMD and other AI hardware suppliers.
Until now, Amazon’s chips have mostly been used inside AWS, where customers access them through cloud services. Selling the chips directly would mark a major shift: Amazon would no longer be only a cloud provider offering AI infrastructure, but also a more active AI chip supplier.
The Move in Brief
| Detail | Update |
|---|---|
| Company | Amazon / AWS |
| Chip line | Trainium |
| Current use | AWS data centers and cloud AI workloads |
| New plan | Sell chips to external companies |
| Main competitor | Nvidia |
| Other rivals | AMD, Google TPUs, custom silicon startups |
| Business goal | Offer lower-cost AI compute alternatives |
| Status | Early talks with potential customers |
From Internal Hardware to External Product
Amazon has spent years building custom silicon for AWS. Its Trainium chips are designed for AI training, while Inferentia chips focus more on AI inference.
The company originally built these chips to reduce dependence on outside suppliers and give AWS customers cheaper AI compute options. Now, Amazon appears ready to turn that internal advantage into a product it can sell more widely.
TechCrunch reported that Amazon’s AI chief Peter DeSantis told Bloomberg that AWS is in talks to sell Trainium chips to other companies for use in data centers. He did not name the potential buyers.
Nvidia Remains Partner and Target
Amazon is not cutting ties with Nvidia.
AWS remains one of Nvidia’s biggest cloud partners. Reuters reported in March 2026 that Nvidia would sell 1 million GPUs to Amazon’s cloud unit by the end of 2027 as part of a major cloud deal.
That makes Amazon’s position interesting. It is both a major Nvidia customer and a growing competitor.
AWS still needs Nvidia GPUs because many AI companies rely on Nvidia’s software ecosystem, especially CUDA, and because demand for AI compute remains extremely high. But by selling Trainium directly, Amazon can offer customers another option, especially for companies looking for better price-performance or more control over their own data center hardware.
The Cost Argument
Amazon’s strongest pitch is likely cost.
Nvidia’s GPUs dominate AI training and inference, but they are expensive and often supply-constrained. AWS has previously positioned Trainium and Inferentia as lower-cost alternatives for customers that want to train or run AI models without paying premium GPU prices.
Reuters reported in 2024 that AWS executives said Amazon’s AI chips could offer 40% to 50% better price-performance in some cases compared with Nvidia-based options.
That argument may become more powerful as AI companies spend billions on compute and look for ways to lower infrastructure costs.
AWS Wants a Bigger Role in AI Infrastructure
Amazon’s AI services are already becoming a meaningful business inside AWS.
Reuters reported in April 2026 that Amazon CEO Andy Jassy said AI services at AWS were generating an annualized revenue run rate of more than $15 billion, equal to roughly 10% of AWS’s $142 billion revenue run rate.
Selling chips directly could create a new hardware revenue stream while also strengthening AWS’s broader AI ecosystem.
It would also help Amazon compete more aggressively with Google, which has been expanding access to its own TPU chips, and Microsoft, which is also developing custom AI silicon.
Big Tech Is Building Around Nvidia, Not Only With Nvidia
Amazon’s move fits a wider industry pattern.
Major Nvidia customers, including Amazon, Google, Microsoft and Meta, are investing in their own AI chips to reduce costs, improve supply control and build more customized AI infrastructure.
Nvidia still leads the market because it offers not only powerful chips but also software, networking, developer tools and a mature ecosystem. But the biggest cloud companies are no longer satisfied with being only buyers.
They want to own more of the AI stack.
A Hard Market to Enter
Selling AI chips directly will not be easy for Amazon.
Nvidia’s strength is not only hardware performance. Its CUDA software ecosystem, developer familiarity and full-stack AI infrastructure make it difficult for rivals to replace.
Customers buying chips for their own data centers also need support, networking, software compatibility, deployment help and long-term hardware roadmaps. Amazon will need to prove that Trainium can work well outside the AWS environment.
That is a different challenge from offering Trainium as a managed cloud service.
Final Take
Amazon’s plan to sell Trainium chips directly shows how serious the AI hardware race has become.
The company still depends heavily on Nvidia, but it also sees a chance to turn its in-house silicon into a broader business. If customers are willing to buy Trainium for their own data centers, Amazon could become a more direct challenger in the AI chip market.
For now, Nvidia remains the clear leader. But Amazon’s move is another sign that the biggest cloud companies want more control over the future of AI compute.