Amazon in Talks to Sell Custom AI Chips to Third Parties in Bid to Undermine Nvidia's Dominance
Miles Bennett
Amazon is in talks with potential clients to sell its custom AI training chip, Trainium, directly into their data centers; this marks the first time Amazon's chip strategy has stepped beyond AWS, taking a direct shot at Nvidia's grip on the AI hardware market.
Who is Amazon selling chips to, and why now?
Amazon AI chief Peter DeSantis confirmed in Paris that the company has begun engaging potential buyers, but did not name specific clients.
His core rationale: demand for locally controlled computing resources outside the US is growing fast.
This means → the "sovereign AI" push in Europe and elsewhere — where data processing must stay in-country — has opened a market gap that AWS cloud services alone cannot fill.
In plain terms = some customers refuse to put data on Amazon's cloud but will buy Amazon's chips to install in their own facilities. Amazon decided to follow that demand.
How is the Trainium chip selling so far?
The third-generation Trainium, which began shipping this year, is "essentially sold out"; the fourth generation, due next year, is already drawing strong interest.
In April Amazon disclosed that Trainium has generated more than $225 billion in revenue commitments.
This means → Trainium is no longer a lab project — it is a mass-produced chip backed by real orders, giving Amazon the confidence to push it beyond AWS.
OpenAI, Anthropic, and Uber have all used Trainium hardware through AWS cloud services.
Won't selling chips cannibalize Amazon's own cloud business?
DeSantis explicitly denied any cannibalization risk, saying: "There's so much unmet demand in AI that I'm not worried about it."
He added that the trend in parts of Europe to reduce reliance on US technology has not hurt AWS business.
In plain terms = Amazon's bet is that the total AI-compute pie is expanding so fast that selling chips and selling cloud are slicing two different pies, not stealing from each other.
What does this mean for Nvidia?
Amazon is moving from "using its own chips inside its own cloud" to "selling chips into other companies' data centers" — stepping directly onto Nvidia GPU's traditional turf.
Since ChatGPT ignited the AI boom, every major cloud giant — Amazon included — has accelerated work on alternatives to Nvidia GPUs.
This reflects a larger shift: hyperscale cloud providers no longer want to be just Nvidia's customers — they aim to become Nvidia's competitors on the chip supply chain itself.
How far does Amazon's chip ambition reach?
Beyond the AI training chip Trainium, Amazon's custom general-purpose processor Graviton is also expanding rapidly.
Over the past three years, Amazon has deployed more Graviton chips in its own computing systems than any other chip type.
Amazon recently began supplying Graviton chips to Meta. This means → Graviton has already proven the "use internally first, then sell externally" playbook — and Trainium is now running the same script.
Content is for reference only, not financial advice.