Amazon Weighs Selling Its AI Chips Directly, Escalating Rivalry with Nvidia

Amazon Web Services is exploring plans to sell its custom-built Trainium artificial intelligence chips directly to other companies for use in their own data centers, a move that would intensify the cloud giant’s competition with Nvidia in the rapidly expanding AI chip market.

The discussions are in early stages, Amazon’s AI chief Peter DeSantis told Bloomberg. A company spokesperson confirmed to Bitcoin World that while AWS has historically declined requests to sell chips directly, CEO Andy Jassy’s recent shareholder letter signaled a potential shift in strategy.

Jassy’s $50 Billion Vision

In his annual shareholder letter published in early April, Jassy wrote that if Amazon’s chip business operated as a standalone company selling to both AWS and third parties, its annual run rate would reach approximately $50 billion. He added that demand for the company’s chips is so strong that selling racks of them to external customers in the future is “quite possible.”

That figure places Amazon’s potential chip revenue on par with Intel’s annual sales. By comparison, Nvidia reported a revenue run rate of roughly $326 billion in its most recent quarter, meaning even a $50 billion competitor would not immediately dethrone the market leader. However, it would mark the most direct challenge yet to Nvidia’s dominance in AI hardware.

Why AWS Has Resisted Selling Chips

For years, AWS has refrained from selling its chips directly, preferring to keep them exclusive to its cloud infrastructure. The reasoning is strategic: AWS profits not only from the compute tokens its chips generate but also from the full ecosystem of services customers need to run AI applications — including storage, security, networking, and monitoring.

Another constraint has been supply. Jassy noted in his letter that capacity for the current Trainium chip sold out almost immediately, as did pre-orders for the next-generation Trainium4, which won’t be available for more than a year. Selling chips externally would risk leaving existing cloud customers waiting unless Amazon can secure additional manufacturing capacity through partners like TSMC, which has recently surpassed Apple as the foundry’s largest customer.

What This Means for the AI Chip Market

If Amazon proceeds, it would enter a market already crowded with well-funded competitors. Nvidia continues to dominate with its GPUs, while AMD and Intel are also vying for a share. Nvidia CEO Jensen Huang recently announced a new $200 billion market opportunity in selling CPUs for AI, further expanding his company’s ambitions into Intel and AMD territory.

Amazon’s move would also signal a broader industry trend: cloud providers are increasingly developing custom silicon to reduce dependence on Nvidia and differentiate their offerings. Google has its Tensor Processing Units, and Microsoft has partnered with AMD to develop its own AI chips.

Conclusion

Amazon’s potential entry into the direct AI chip market represents a significant escalation in the battle for AI hardware supremacy. While the company faces manufacturing constraints and the risk of cannibalizing its cloud services revenue, the prospect of a $50 billion chip business — backed by the engineering and logistical power of one of the world’s largest companies — could reshape the competitive landscape. For now, the talks remain exploratory, but Jassy’s public comments suggest the idea has serious executive backing.

FAQs

Q1: What is Amazon’s Trainium chip?
Trainium is Amazon Web Services’ custom-designed AI chip optimized for training machine learning models. It is currently available only within AWS’s cloud infrastructure.

Q2: How would selling chips directly affect AWS customers?
If Amazon sells chips to third parties, it could create supply constraints for existing AWS cloud customers unless manufacturing capacity increases. The company would need to balance external sales with internal demand.

Q3: Is Amazon likely to succeed in challenging Nvidia?
Amazon’s potential $50 billion chip revenue would be significant but still far below Nvidia’s current run rate. Success would depend on manufacturing capacity, software ecosystem support, and the ability to match Nvidia’s performance and developer tools.