Amazon Wants to Sell You Chips, But Not the Kind You Think
Nvidia has a chokehold on the tech industry. We all know it. Every major cloud provider is currently paying the Jensen Huang tax, handing over billions of dollars for H100 and Blackwell GPUs just to stay in the artificial intelligence race. But Amazon is tired of playing defense.
According to a new report from TechCrunch, Amazon Web Services is in talks to sell its custom AI silicon directly to other data centers. Yes, you read that right. AWS wants to sell its Trainium and Inferentia hardware to the very companies it usually tries to crush in the cloud market. It is a wild, counterintuitive move.
And it might be the smartest decision Andy Jassy has made since taking the crown.
The Fifty Billion Dollar Question
Andy Jassy recently teased that this hardware play represents a 50 billion dollar opportunity for Amazon. That is a massive number. To put it in perspective, AWS is already the undisputed king of cloud infrastructure, but Jassy thinks selling chips to outsiders can supercharge their growth. But let's be honest about the hurdles here.
The reality is that nobody actually wants to buy hardware from their biggest competitor. If you're a mid-sized data center operator, buying Trainium chips from AWS feels a bit like buying groceries from the restaurant that's trying to put you out of business. You're funding your own demise.
Yet, the math might force their hand.
Nvidia chips are expensive. They're also incredibly hard to get. If AWS can offer comparable performance at a fraction of the cost, and actually deliver the silicon without a nine-month waiting list, pragmatism will win. Companies will swallow their pride. They'll sign the checks.
"We want to give people choice," Jassy has said, though the subtext is clear: we want to stop paying Nvidia, and we want you to stop paying them too.
Here's What Most Coverage Misses
Most analysts are treating this as a simple hardware play. It's not. This is a software trap.
Nvidia didn't win the AI race just because its silicon is fast. It won because of CUDA, the software platform that developers have spent a decade learning. If you write code for Nvidia chips, it just works. It's an incredibly sticky ecosystem. If you want to use another chip, you have to rewrite your software, and nobody wants to do that.
Amazon has its own software layer called Neuron. It's fine, but it's not CUDA. By selling Trainium chips to other data centers, Amazon is trying to force Neuron into the wild. They need to build an ecosystem. If they can get thousands of developers outside of AWS using Neuron, they break Nvidia's software monopoly. That's the real prize.
So, will it work?
It's a massive gamble. Microsoft and Google are also building their own chips, the Maia and TPU lines. But they're keeping them locked inside their own cloud kingdoms for now. Amazon is the only one crazy enough to open the gates and sell the physical silicon to the public. It's risky, but staying quiet and letting Nvidia dictate the margins of the entire tech sector is even riskier. We're about to find out if the market is ready to buy its weapons from the biggest empire in the cloud.
Frequently Asked Questions
Why is Amazon selling its AI chips to other companies?
Amazon wants to break Nvidia's monopoly on AI hardware and establish its own software ecosystem, Neuron, as an industry standard. CEO Andy Jassy believes this merchant silicon market represents a 50 billion dollar opportunity for AWS.
What are the names of Amazon's AI chips?
Amazon designs two main families of custom silicon: Trainium, which is optimized for training large language models, and Inferentia, which is designed to run those models once they are trained.
Will competitors actually buy hardware from AWS?
Some will hesitate because AWS is their direct cloud competitor. However, because Nvidia GPUs are incredibly expensive and suffer from severe supply shortages, many data center operators will likely buy Amazon's chips to save money and get hardware faster.