Andy Jassy just floated the idea of selling Amazon’s custom AI chips to other companies. You know, the same chips Amazon built specifically because it didn’t want to keep paying Nvidia’s premium prices. The same chips designed to give AWS a competitive edge in the cloud AI wars. Now they might sell them to you.
Let me get this straight: Amazon spent years and untold millions developing Trainium and Inferentia chips to reduce dependence on external suppliers. They’ve been pushing AWS customers hard to adopt these chips instead of renting Nvidia GPUs. And now, with their AI business hitting a $15 billion annual run rate as of Q1 2026, Jassy says there’s “so much demand” they might sell racks of these chips to third parties.
The Audacity Is Almost Impressive
This is classic Amazon. Build something for internal use, realize you can monetize it, then turn your competitive advantage into a product. It worked with AWS itself—Amazon needed servers for its retail business, built a massive infrastructure, then decided to rent it out. That turned into the most profitable part of the company.
But selling AI chips is a different beast. When Amazon launched AWS, it wasn’t directly competing with its potential customers. Now? Every company buying these chips would be funding the infrastructure of a competitor that operates in nearly every sector imaginable. Amazon sells groceries, streams video, runs logistics networks, and hosts half the internet. You’d be paying them to give you the tools to compete with them.
Nvidia and AMD Should Be Sweating
For Nvidia and AMD, this is a nightmare scenario wrapped in a press release. Nvidia has owned the AI chip market with margins that would make a luxury watchmaker blush. AMD has been fighting for scraps, trying to position itself as the value alternative. Now here comes Amazon with chips that are already deployed at scale, battle-tested in production environments, and backed by the largest cloud provider on the planet.
Google tried this strategy first with its TPU chips, and it’s worked reasonably well for them. But Google doesn’t have AWS’s market share or Amazon’s willingness to undercut everyone on price. If Amazon decides to sell these chips, you can bet they’ll be priced to move.
The Real Question Nobody’s Asking
Why would Amazon do this? The obvious answer is money. That $15 billion run rate is nice, but it’s mostly from AWS customers using Amazon’s chips in the cloud. Selling physical hardware opens up a new revenue stream from companies that want to run AI workloads on-premises or in their own data centers.
The less obvious answer? Market control. If Amazon can get its chips into enough companies, it creates a moat. Those companies build their AI infrastructure around Amazon’s architecture. They train their teams on Amazon’s tools. They optimize their models for Amazon’s specifications. And when they need to scale or add cloud capacity, guess which cloud provider has perfect compatibility?
What This Means for AI Tool Buyers
If you’re shopping for AI infrastructure right now, this announcement changes the calculation. Nvidia’s H100s are still the gold standard, but they’re expensive and hard to get. AMD’s MI300 series offers solid performance at better prices. And now Amazon might enter the ring with chips that have been running real AI workloads at massive scale.
The smart move? Wait and see what Amazon actually delivers and at what price point. Don’t get caught up in the hype of a “maybe we’ll sell them” statement from a CEO defending aggressive AI spending to nervous investors. Jassy needs to show that all this investment will pay off, and floating the idea of a new revenue stream is an easy way to calm shareholder concerns.
But if Amazon does follow through, the AI chip market is about to get a lot more interesting. And by interesting, I mean cheaper for buyers and more painful for Nvidia’s profit margins. That’s the only part of this story I’m genuinely excited about.
đź•’ Published: