Amazon Wants to Sell Its AI Chips to Rival Data Centers—Here’s the Real Play

Amazon is finally getting serious about challenging Nvidia in the AI chip game. Not just by building its own silicon for AWS, but by selling those chips to other data centers. That’s a big shift.

CEO Andy Jassy recently said this represents a $50 billion opportunity for the company. That’s a number that makes you sit up straight. But let’s be real—this isn’t going to be easy.

AWS has been making its own chips for a while now. Trainium for training, Inferentia for inference. They’re decent, but they haven’t exactly set the world on fire. The reason is simple: Nvidia’s CUDA ecosystem is a monster. Once you’re locked into CUDA, switching to something else is painful. Developers don’t want to rewrite their code, and the performance gap isn’t big enough to justify the hassle.

But here’s where Amazon might have an angle. They’re not just selling chips; they’re selling access to their entire cloud infrastructure. If you’re a data center operator, buying AWS chips means you can potentially offer your customers seamless integration with AWS services. That’s a compelling pitch for some workloads, especially if you’re already running a hybrid setup.

Still, I’m skeptical about the $50 billion figure. That’s a lot of chips. Nvidia’s data center revenue alone was over $47 billion last quarter. Amazon would need to capture a meaningful chunk of that market, and they’re starting from pretty much zero outside their own walls.

The real question is whether Amazon can create a software ecosystem that developers actually want to use. They’ve tried with Neuron, their SDK for Trainium and Inferentia, but it’s nowhere near as mature as CUDA. And developers hate dealing with immature toolchains.

My take: This is Amazon playing the long game. They don’t need to beat Nvidia overnight. They just need to be good enough for enough customers to make the switch. If they can offer better pricing or unique features—like tighter integration with AWS services—they might carve out a niche. But $50 billion? That’s a stretch, at least for the next few years.

What’s more interesting to me is the broader trend. Hyperscalers like Amazon, Google, and Microsoft are all building their own chips now. They’re tired of paying Nvidia’s margins. This is a structural shift in the industry, and Nvidia will eventually feel the pressure. But it’s going to take time, and Amazon’s move to sell chips externally is just one piece of that puzzle.

Let’s see if they can actually execute. The hardware is solid, but the software story needs a lot of work.