Amazon Web Services is considering a more direct step into the AI chip market: selling its own Trainium chips to other companies for use in data centers. The talks are still in the early stages, but the idea marks a notable shift for a cloud business that has mostly kept its homegrown chips inside AWS.
The potential move matters because Nvidia remains the dominant force in AI chips. AWS already competes by offering customers access to its own chips through the cloud, but direct chip sales would place Amazon closer to Nvidia’s core business.
What AWS Is Exploring
Amazon’s AI chief Peter DeSantis told Bloomberg that AWS is in talks to sell Trainium to other companies for data centers. DeSantis did not identify which companies might buy the chips.
TechCrunch reported that the discussions are early. The idea is tied to Amazon CEO Andy Jassy’s annual shareholder letter in early April, where he described unusually strong demand for Amazon’s own AI chips and raised the possibility of selling them beyond AWS.
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. There’s so much demand for our chips that it’s quite possible we’ll sell racks of them to third parties in the future.
AWS spokesperson Doron Aronson also confirmed the possibility. He said, “While we’ve historically declined requests to sell chips directly, Andy noted it’s quite possible we’ll sell racks of them to third parties in the future.”
Why This Would Be A Bigger Nvidia Challenge
AWS already has a chip strategy, but it has mainly been a cloud strategy. Customers use Amazon’s chips through AWS services, rather than buying the hardware for their own data centers.
That model gives AWS more than one way to make money. The company can charge for the AI tokens processed on its cloud. It can also sell the surrounding services that AI applications need, including storage, security, networking, and monitoring services.
Selling Trainium racks to third parties would be different. It would mean AWS is not only using chips to support its cloud, but also offering them as products in a market where Nvidia has built a major position.
The scale Jassy described is significant. A $50 billion competitor would not necessarily overturn Nvidia’s position, especially when Nvidia is currently on a $326 billion revenue run rate if it keeps delivering quarters like the last one. But TechCrunch notes that the figure is akin to Intel’s annual revenues, which makes Amazon’s potential chip business large enough to matter.
The Supply Problem
The largest question is whether AWS can sell chips externally without limiting supply for its own customers. The source article makes clear that demand is already tight.
In the same shareholder letter in April, Jassy said current Trainium chip capacity had sold out almost instantly. He also said capacity for Trainium4 had sold out too, even though Trainium4 will not be available for more than a year.
That creates a practical constraint. If AWS sells chips to other companies, it may have to leave existing customers waiting unless it can produce more chips than it needs for its own cloud business.
The source also points to manufacturing pressure. AWS would need help from manufacturing partners such as TSMC. But it would have to compete for capacity in a market where Nvidia has recently supplanted Apple to become TSMC’s largest customer.
What Amazon Gains And What It Risks
The logic of keeping Trainium inside AWS is clear. A customer using AWS chips may also need multiple AWS services around the AI workload. That creates a broader commercial relationship than a one-time or hardware-focused chip sale.
Direct sales could still give Amazon a new way to participate in AI infrastructure demand. If companies want Trainium for their own data centers, AWS could meet demand beyond the boundaries of its cloud platform.
But the tradeoff is also clear from the source. Every rack sold to a third party is hardware that may not be available to AWS customers, unless production expands enough to cover both paths. That is why supply, not only demand, is central to the question.
- Current model: AWS keeps Trainium mainly inside its cloud and sells services around it.
- Possible new model: AWS sells racks of Trainium chips to third parties.
- Main constraint: Trainium capacity has been selling out faster than AWS can produce it.
- Competitive context: Nvidia remains far larger, but Amazon’s stated chip run rate would still be substantial.
The Broader AI Chip Race
The source places Amazon’s ambitions alongside Nvidia’s own expansion plans. Nvidia founder and CEO Jensen Huang recently said he has found a brand new $200 billion market for Nvidia in selling CPUs for AI, not just GPUs. That move pushes Nvidia into territory associated with Intel and AMD.
Amazon’s possible Trainium sales point in the opposite direction: a cloud company moving closer to the chip market. If AWS follows through, it would not simply be offering another way to run AI workloads in the cloud. It would be putting Amazon-built AI chips into other companies’ data centers.
For now, the talks remain early, and AWS has not named potential buyers. The most important facts are already visible: Amazon sees major demand for its AI chips, Nvidia’s market is large enough to attract deeper competition, and supply may decide how far AWS can go.