Amazon is considering expanding the sale of its self-developed AI chips from its internal cloud services to external markets. As AWS continues to invest heavily in AI infrastructure, this move suggests that Amazon's competition with Nvidia may extend beyond cloud computing services to include data center chip supply.

AWS is in talks to sell Trainium.

According to Bloomberg, citing Peter DeSantis, head of AWS AI, AWS is in talks with several companies to sell Trainium chips for data center deployments. However, he did not disclose the list of potential customers.

AWS has not previously sold these types of chips directly to external customers. Amazon's more common practice is to use its self-developed chips in its own cloud platform and then sell computing power and related services to customers.

Cloud services remain the primary monetization method.

For AWS, chips themselves are not the only source of revenue. When enterprises run AI applications in the cloud, in addition to purchasing computing power, they also need supporting services such as storage, security, networking, and monitoring. This is also an important reason why AWS has historically preferred to keep chips within its own platform.

However, Amazon management has begun publicly discussing the possibility of selling to external parties. AWS spokesperson Doron Aronson also confirmed that while the company has typically refused to sell chips directly, selling entire cabinets of equipment to third parties in the future is not out of the question.

Tight production capacity remains a constraint.

One real problem facing Amazon is that supply remains tight. The company previously disclosed that existing Trainium production capacity was almost sold out quickly, and even the next-generation Trainium 4, which has not yet been officially launched, has been fully booked in advance.

This means that if AWS were to expand its external sales now, it might need to make a trade-off between meeting the needs of existing cloud customers and expanding into external markets. Unless its foundry partners can significantly increase production capacity, new business will be difficult to scale up quickly.

From a manufacturing perspective, AWS still needs to compete with Nvidia for advanced production capacity. The report mentions that Amazon's chip production relies on partners such as TSMC, and Nvidia is currently one of TSMC's most important customers.

The $50 billion target is aimed directly at Nvidia.

DeSantis stated that if the AWS chip business is considered a separate business, and the chips produced this year are sold to both AWS and external customers, its annualized revenue could reach approximately $50 billion.

This figure is still significantly lower than Nvidia's current annualized revenue of approximately $326 billion, but it's enough to demonstrate Amazon's desire for a more direct voice in the AI chip market. The report suggests this will put more direct competitive pressure on Nvidia than ever before from AWS.