Amazon Web Services is exploring direct sales of its custom Trainium AI chips to third-party companies, a move that would intensify competition with Nvidia in the $200 billion-plus market for artificial intelligence hardware. AWS chief Peter DeSantis told Bloomberg the cloud giant is in early-stage talks with potential buyers, though he declined to name specific companies.
The talks stem from Amazon CEO Andy Jassy’s annual shareholder letter in early April, in which he wrote: “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.”
Also read: AI inference startup Baseten reportedly raising $1.5B at $13B valuation months after last mega round
Why AWS is reconsidering its chip sales strategy
AWS has historically resisted selling Trainium chips directly, despite repeated requests from customers. The reason is financial: the company profits not just from the AI tokens processed on its chips, but from the full stack of cloud services — storage, security, networking, and monitoring — that companies bundle with AWS compute capacity. Selling chips directly would cut into that bundled revenue stream.
AWS spokesperson Doron Aronson confirmed the shift in strategy. “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,” Aronson said.
The manufacturing bottleneck
The biggest obstacle to Amazon’s chip ambitions is manufacturing capacity. Jassy stated in the same shareholder letter that current Trainium capacity sold out almost instantly, and the same is true for the next-generation Trainium4 — which won’t be available for more than a year. This was before AWS formally added OpenAI’s models to its cloud offerings, further straining demand.
To sell chips externally, Amazon would need to secure additional production capacity from its manufacturing partner TSMC, which has recently supplanted Apple as the foundry’s largest customer. Competing with Nvidia for TSMC’s advanced packaging capacity would be a significant challenge.
What this means for Nvidia
A $50 billion Amazon chip business would not single-handedly displace Nvidia, which reported a $326 billion annualized revenue run rate in its most recent quarter. However, it would position AWS at a scale comparable to Intel’s annual revenue, creating a credible third force in AI hardware alongside Nvidia and AMD.
Nvidia CEO Jensen Huang recently announced a new $200 billion market opportunity selling CPUs for AI workloads, moving into territory traditionally held by Intel and AMD. Jassy’s $50 billion ambition represents a direct response — pushing AWS deeper into Nvidia’s core GPU market rather than ceding the cloud AI hardware space.
Frequently Asked Questions
What are Amazon’s Trainium chips?
Trainium are custom AI accelerator chips designed by Amazon Web Services to train and run machine learning models, competing directly with Nvidia’s GPUs in the cloud computing market.
Why has AWS previously refused to sell its chips directly?
AWS historically kept Trainium exclusive to its own cloud infrastructure because the real profit comes from bundling chip access with storage, security, networking, and monitoring services — creating a waterfall revenue effect.
How big could Amazon’s chip business become?
CEO Andy Jassy estimated in his April 2025 shareholder letter that if the chip division were a standalone business, its annual run rate would be approximately $50 billion, comparable to Intel’s yearly revenue.
What challenges does Amazon face in selling chips externally?
Current Trainium capacity is already sold out through the next generation (Trainium4), and Amazon would need to secure additional manufacturing capacity from TSMC — which has recently surpassed Apple as TSMC’s largest customer.
Be the first to comment