3 Sources
[1]
Amazon hopes to challenge Nvidia more directly by selling its AI chips
If Amazon Web Services has its way, the cloud giant is going to push even deeper into Nvidia's market, in what might be one of the biggest challenges to Nvidia's AI chip dominance we've seen so far. Amazon's AI chief Peter DeSantis told Bloomberg that AWS is in talks to sell its AI chip Trainium
[2]
Amazon in talks to sell custom AI chips in bid to cut Nvidia's dominance
Amazon is planning to sell its custom artificial intelligence chips to other companies. This move aims to compete with Nvidia's strong hold in the market. Amazon's AI chief confirmed discussions are underway. The company sees a growing demand for AI infrastructure globally. This expansion could
[3]
Inside the lab where Amazon is taking on Nvidia: Jon Erlichman
In an industrial park in Austin, Texas, behind glass doors that look like any other corporate office, Amazon is building one of the most important businesses in technology. Some investors haven't noticed yet. But they will soon. We spent a day inside Annapurna Labs, where Amazon designs
Share
Copy Link
Amazon Web Services is in talks to sell its Trainium AI chips to other companies for use in data centers, marking a major shift in strategy. The move could create a $50 billion business that directly competes with Nvidia's AI chip dominance. AWS AI chief Peter DeSantis confirmed early-stage discussions are underway, driven by surging demand that has seen Trainium3 capacity sell out almost instantly.
Amazon Web Services is preparing to challenge Nvidia's dominance in the AI chip market by selling its custom AI chips directly to other companies for use in data centers
1
. Peter DeSantis, Amazon's AI chief, confirmed that discussions are underway with potential customers, though he declined to name specific companies involved in the talks2
. The initiative stems from Amazon CEO Andy Jassy's April shareholder letter, where he revealed that if the chips business operated as a standalone entity, it would generate an annual run rate of approximately $50 billion1
.
Source: TechCrunch
The decision to potentially sell Amazon AI chips externally comes as demand for AI infrastructure continues to surge globally. Trainium, Amazon's AI accelerator introduced in 2020, has already secured marquee buyers including OpenAI, Anthropic, and Uber Technologies, who access the hardware through Amazon Web Services
2
. The chip has generated more than $225 billion in revenue commitments as of April. DeSantis noted that Trainium3, which began shipping earlier this year, is "largely sold out," while capacity for Trainium4, expected to debut next year, has already been claimed despite being over a year away from availability1
2
.The development of these custom AI chips happens at Annapurna Labs in Austin, Texas, where Amazon designs cost-effective processors for large language models and AI workloads
3
. Amazon acquired the small Israeli chip design firm in 2015 for a reported $350 million, a deal that barely registered in tech news at the time but has since proven to be one of Amazon's smartest acquisitions3
. The custom chip business now runs at more than $20 billion annually, growing at triple-digit rates3
. Engineers at Annapurna have created three chip families: Graviton for general-purpose processing, Nitro for cloud infrastructure, and Trainium for AI training and inference. The latest Trainium3 chip delivers 4.4 times the performance of its predecessor by shrinking silicon technology from 5 nanometers to 3 nanometers3
.
Source: BNN
Related Stories
While Nvidia currently operates on a $326 billion revenue run rate, a $50 billion competitor would represent a significant challenge comparable to Intel's annual revenues
1
. The move mirrors Google's strategy, as Alphabet CEO Sundar Pichai announced in April that Google would deliver its tensor processing units to a select group of customers for use in their own data centers. DeSantis indicated that growing demand for sovereign AI services in Europe and other regions, where computing resources are controlled locally and subject to local laws, is partly driving the decision to sell Trainium outside of AWS. He dismissed concerns that selling chips directly would cannibalize cloud computing business, stating "there's so much underconsumption in AI"2
.Selling chips externally presents logistical challenges for Amazon, as it would need to manufacture surplus capacity through partners like TSMC while avoiding leaving current AWS customers on waiting lists
1
. This becomes more complex given that Nvidia has recently supplanted Apple to become TSMC's largest customer1
. Historically, AWS has resisted selling chips because its revenue model relies on a waterfall effect—charging not just for AI tokens processed but also for storage, security, networking, and monitoring services that accompany cloud usage1
. AWS spokesperson Doron Aronson confirmed the potential shift, noting that while the company has historically declined requests to sell chips directly, Andy Jassy indicated it's quite possible they'll sell racks to third parties in the future1
. Amazon shares gained as much as 1.8% on the news, reaching an intraday high of $241.822
. As Jassy wrote in his shareholder letter, "Our chips business is on fire"3
.Summarized by
Navi
1
Technology

2
Technology

3
Science and Research
