Amazon targets Nvidia with new push to sell its AI chips
▼ Summary
– AWS is in early talks to sell its Trainium AI chips to other companies for data center use, challenging Nvidia’s dominance.
– Amazon CEO Andy Jassy’s shareholder letter indicated potential chip sales, estimating a $50 billion annual run rate if sold as a standalone business.
– AWS has historically resisted selling chips due to revenue from associated services like storage and security, creating a waterfall effect.
– Current Trainium chip capacity is sold out, including the upcoming Trainium4, making sales to others difficult without manufacturing surpluses.
– Amazon’s potential $50 billion chip market would directly compete with Nvidia, which has a $326 billion revenue run rate, while Nvidia targets a new $200 billion CPU market.
Amazon Web Services is quietly laying the groundwork for a direct assault on Nvidia’s dominance in the AI chip market. According to AWS AI chief Peter DeSantis, the cloud giant is in early-stage discussions to sell its proprietary Trainium AI chips to external companies for use in their own data centers. While DeSantis declined to name potential buyers, the move signals a significant strategic shift for the company.
These talks, which Amazon confirms are still nascent, trace back to CEO Andy Jassy’s April shareholder letter. In it, Jassy hinted that demand for Amazon’s homegrown chips was so intense he was considering selling them externally. “If our chips business was a standalone entity, 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,” Jassy wrote. “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.”
How realistic is a $50 billion competitor to Nvidia? It wouldn’t cripple the chip giant, which currently runs on a $326 billion revenue run rate if it sustains its recent quarterly performance. But it would put Amazon’s chip business on par with Intel’s annual revenues.
For years, AWS resisted selling its AI chips for a simple reason: the real money isn’t in the silicon itself. The company profits from a waterfall effect , charging customers for AI tokens processed on its cloud, plus a suite of ancillary services like storage, security, networking, and monitoring. Selling chips externally would disrupt that model.
Capacity constraints also play a role. In his April letter, Jassy noted that current Trainium chip capacity sold out almost instantly, as did the capacity for the next generation, Trainium4, which won’t be available for over a year. This was before AWS even added OpenAI’s models to its roster. Selling chips to third parties would likely force existing customers onto waiting lists, unless Amazon can somehow manufacture a surplus through partners like TSMC. But TSMC’s foundry capacity is already stretched thin, and Nvidia recently overtook Apple as the foundry’s largest customer.
AWS spokesperson Doron Aronson confirmed the shift: “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.”
The timing is telling. Nvidia CEO Jensen Huang recently announced a $200 billion new market for selling CPUs for AI, moving into Intel and AMD territory. Jassy’s ambition is equally bold: a $50 billion chip business that would directly challenge Nvidia’s stronghold. The battle for AI silicon supremacy is only heating up.
(Source: TechCrunch)
