Amazon's Bold Move: Challenging Nvidia with AI Chips Sales

TL;DR

• Amazon Web Services is now exploring direct sales of its Trainium AI chips to outside customers, a move that would push it farther into Nvidia’s core market.
• CEO Andy Jassy says Amazon’s chip business already has an annual run rate above $20 billion inside AWS and could be worth about $50 billion if sold more broadly.
• The strategy could reshape cloud computing and data center buying, but Nvidia still dominates AI chips and Amazon’s chips remain a scaling challenge.

AWS Takes a Bigger Swing at Nvidia

Amazon is moving from an inside-the-cloud chip strategy to a more direct hardware business, with AWS now considering selling its Trainium AI chips to other companies for use in their own data centers. The shift would put Amazon in more direct competition with Nvidia, whose GPUs have become the default choice for training and running many AI systems.

The idea is still early, but it marks a notable change in Amazon’s posture. AWS had historically declined requests to sell chips directly, but Amazon officials now say selling racks of chips to third parties is “quite possible” in the future.

Why Amazon Thinks the Market Is Huge

Amazon CEO Andy Jassy has been unusually explicit about the opportunity. In his annual shareholder letter, he said Amazon’s homegrown chips are so in demand that, if the chip business were a standalone company selling to AWS and external customers, it could generate an annual run rate of about $50 billion.

That estimate builds on Amazon’s current internal scale. Jassy said the company’s chip division already exceeds a $20 billion annual run rate, with triple-digit year-over-year growth, though that figure reflects revenue mostly tied to AWS usage rather than an independent chip business. The implication is that Amazon believes it is only monetizing a fraction of its chip demand today.

What Amazon Is Selling

Amazon’s custom silicon effort includes Trainium for AI training, Inferentia for inference, and other in-house hardware used across AWS infrastructure. The company has positioned these chips as lower-cost alternatives to Nvidia GPUs, arguing they can improve price-performance for certain workloads.

Jassy said Trainium2 offers roughly 30% better price-performance than comparable GPUs, and Trainium3 is already shipping with further gains. AWS also says Bedrock, its managed AI service, already runs the majority of its inference on Trainium.

The Nvidia Challenge Is Real, but Uneven

Amazon’s push is significant, but it is not the same as overtaking Nvidia. AWS executives have said the goal is to offer customers more choice, not necessarily to “unseat Nvidia,” and they have described the market as large enough for multiple vendors.

At the same time, Amazon has faced adoption hurdles. Earlier reporting described low usage rates for Trainium and Inferentia among AWS’s largest customers, along with compatibility and migration issues that could limit broader uptake. That suggests Amazon’s chips have gained traction inside AWS, but expanding them into a standalone external business will require much more than price competitiveness alone.

Why Data Centers Matter

If Amazon starts selling Trainium directly, the biggest immediate impact may be in data center procurement. Enterprises, sovereign cloud operators, and AI startups could buy integrated racks or chip clusters instead of depending entirely on Nvidia-based systems from cloud providers.

That could matter especially in regions and industries that want more control over AI infrastructure, lower costs, or local deployment options. It also reflects a broader shift in the AI market: hyperscalers are increasingly designing their own silicon to reduce dependence on external chip suppliers and lower long-term infrastructure costs.

The Economics Behind the Push

Amazon argues that custom chips can save it “tens of billions” in capital expenditures annually at scale and improve operating margins compared with relying on third-party chips for inference. That kind of internal savings helps explain why AWS is investing so heavily in its own silicon stack.

There is also a strategic business angle. Selling chips externally would create a new revenue stream, but it could also deepen AWS’s role as a one-stop AI infrastructure provider spanning networking, training, inference, and cloud services. In other words, Amazon is not just trying to sell chips; it is trying to sell an AI platform.

What Happens Next

For now, the chip-selling effort is in early discussions, so Amazon is not yet launching Trainium as a broad standalone product line. Even so, the direction is clear: AWS wants its custom silicon to matter beyond its own cloud, and that puts it on a more direct collision course with Nvidia.

If Amazon can turn Trainium into a real external business, the result could be a major new competitive front in AI infrastructure. If it cannot, the chips may still remain valuable as a way to make AWS cheaper and more efficient than rivals.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.