Amazon Signals a Direct Push Into Nvidia’s Turf With Outside Sales of Trainium Chips

Amazon Web Services is weighing a move that could reshape the economics of artificial intelligence infrastructure: selling its homegrown Trainium chips to outside customers, not just using them internally for AWS workloads. If the plan advances, Amazon would stop being only a cloud provider with proprietary silicon and begin acting more like a full-fledged chip vendor — a step that would put it in more direct competition with Nvidia than ever before.

That possibility was outlined by Amazon’s AI chief Peter DeSantis in remarks to Bloomberg, where he said the company is in discussions about offering Trainium chips to other firms for deployment in their own data centers. Amazon later told TechCrunch that those conversations are still in an early phase, but executives have not ruled out the idea. The shift follows a strikingly candid message from CEO Andy Jassy in his annual shareholder letter, where he argued that demand for Amazon’s AI chips is strong enough that the business could eventually generate tens of billions of dollars a year if treated as a standalone operation.

The prospect matters because Nvidia remains the dominant supplier of the accelerators that power most AI training and inference systems. Amazon’s entry into the market would not instantly displace that dominance, but it could carve out a meaningful alternative for companies looking to diversify their hardware supply. It would also deepen a long-running race among hyperscalers to control the stack that underpins generative AI — from chips to model hosting to storage, networking, and security.

Amazon’s chip strategy is moving beyond self-use

For years, Amazon’s custom chips were primarily part of an internal strategy: reduce dependence on outside vendors, lower costs, and create infrastructure tailored to AWS customers. That approach has made Amazon one of the few cloud giants able to challenge Nvidia’s pricing power while still keeping its own cloud margins intact.

Now the company appears to be exploring a more ambitious route. Rather than reserving Trainium solely for AWS capacity, Amazon is considering whether customers could buy the chips directly and install them in their own facilities. That would place Amazon closer to the model used by traditional semiconductor companies, even if the sales are still tightly linked to AWS services and manufacturing partners.

According to the reporting, the commercial discussions are not yet a finalized product launch. Amazon says the idea is still in early stages. But the company’s leadership has been unusually open about the scale of demand it sees for its AI silicon, and that transparency suggests the concept is being treated as more than a speculative thought experiment.

What Amazon says is driving the idea

Jassy’s shareholder letter in April laid much of the groundwork for the current discussion. In it, he said that if Amazon’s chip business were separated from the rest of AWS and sold to both AWS and third parties, it could produce an annual run rate of roughly $50 billion. He also said demand was so strong that Amazon might eventually sell racks of chips rather than only using them in its own cloud.

That estimate is important for two reasons. First, it reveals that Amazon sees its chips as a business in their own right, not merely a cost-saving tool. Second, it suggests the company believes the addressable market could be large enough to justify taking on some of the complexities that come with becoming a chip supplier.

Jassy has argued that if Amazon’s chip operation were treated as a separate business serving AWS and external buyers, it could eventually produce about $50 billion in annual run-rate revenue.

That figure does not mean Amazon would suddenly rival Nvidia in raw size. But it would be a formidable business by almost any standard, and especially notable in a market where supply constraints and long-term AI demand have made chip capacity one of the most valuable resources in tech.

Why AWS has held back until now

There are strong reasons Amazon has historically avoided selling chips directly. The most obvious is strategic: AWS does not simply earn money by processing AI workloads. The company also benefits from an ecosystem of adjacent services that customers need to run those workloads reliably at scale.

When a customer uses AWS hardware for AI, Amazon earns revenue not only from the processing itself but also from storage, networking, observability, security, and management tools. That makes the cloud business far more profitable than a pure hardware sale would be. Selling chips separately could mean giving up some of that “waterfall” of downstream revenue.

There is also the issue of supply. Amazon has said its chip capacity is already effectively sold out. Jassy said in April that current Trainium capacity was snapped up almost immediately, and that future capacity for Trainium4 had also been reserved even though the chip will not be available for more than a year. In other words, Amazon does not appear to have excess inventory sitting idle; it appears to have the opposite problem.

The supply challenge

If Amazon begins shipping chips to outside buyers, it would need to decide whether to hold back internal capacity, expand production, or both. Any move into third-party sales could force AWS to manage customer waiting lists more carefully, unless manufacturing partners can deliver a significant increase in output.

That is where foundry capacity becomes a major constraint. Amazon relies on external manufacturing partners such as TSMC, the world’s leading chip foundry. But demand for advanced semiconductor production is fierce, and the competition for wafer starts is intense. Nvidia remains one of TSMC’s largest customers, and Amazon would need to compete not only with Nvidia’s purchase orders but also with the broader demand coming from Apple, AMD, and other giants that depend on the same fabrication ecosystem.

Any meaningful scale-up would therefore require Amazon to do more than design good chips. It would need a dependable supply chain, long-term manufacturing commitments, and a go-to-market strategy that can persuade companies to buy Amazon silicon instead of remaining in Nvidia’s orbit.

How this could change the AI hardware market

The AI chip market is already undergoing one of the fastest industrial shifts in modern tech history. Nvidia has become the defining winner, with its GPUs serving as the backbone of training and inference for major frontier models. But the market is also broadening. Google has its TPUs, Microsoft has explored custom silicon, and Amazon has been steadily building its own alternatives through Trainium and Inferentia.

What makes Amazon’s possible move different is that it would put the company more squarely in the business of selling hardware to other enterprises. That would create a new layer of competition: not just cloud versus cloud, but chip vendor versus chip vendor.

For customers, that could bring more choice. Some companies may prefer a non-Nvidia option if it offers better pricing, sufficient performance, or deeper integration with AWS tooling. For Amazon, the benefit would be strategic leverage: a way to win business even from customers who are not fully committed to AWS cloud services, while still promoting an ecosystem that keeps them tied to Amazon’s operational stack.

What a $50 billion chip business would mean

Amazon’s suggested run-rate is worth unpacking. A business at that scale would be enormous in its own right. It would not necessarily match Nvidia’s current trajectory, but it would place Amazon’s chips in the same league as some of the largest semiconductor companies in history.

For context, the source material notes that a $50 billion annual business would approach Intel-like scale. That comparison matters because Intel’s revenues have long represented the standard for a major standalone chip company. If AWS can build a chip operation of that size while still remaining a cloud giant, it would represent a structural change in how infrastructure providers compete in AI.

The implication is broader than revenue. A large enough Amazon chip business could influence pricing across the entire AI stack, alter procurement decisions at enterprise data centers, and force Nvidia to defend more aggressively against price-sensitive buyers.

Nvidia is still the benchmark Amazon has to beat

Despite the attention around Amazon’s ambitions, Nvidia remains far ahead. The company’s current revenue run rate is roughly $326 billion, based on the figures referenced in the source material. That means Amazon’s projected $50 billion chip business would not come close to overtaking Nvidia on size alone.

But the relevant question is not whether Amazon can instantly dethrone Nvidia. It is whether Amazon can create a credible alternative at scale. In markets like AI infrastructure, even a well-funded second source can matter enormously. Enterprise buyers often want redundancy, price leverage, and supply diversification. If AWS can offer a high-performance chip path with robust software support, it could become a serious option for certain classes of workloads.

Amazon also has a built-in advantage: it already serves a massive customer base through AWS. That means it has direct channels to large buyers, deployment experience at scale, and the infrastructure needed to bundle chips with higher-level services. Nvidia’s business is powerful, but it is not integrated into a cloud platform the way Amazon’s is.

Amazon’s edge is not just silicon

The company’s value proposition would likely extend beyond the chip itself. Amazon can bundle compute with storage, networking, compliance, security, monitoring, and deployment tools. That makes the chip a gateway to an environment rather than a standalone product.

That bundling power is a major reason Amazon’s AI strategy is so significant. Even if Trainium chips do not match Nvidia’s GPUs in every benchmark, AWS can potentially win customers with lower total cost of ownership, easier integration, and enterprise-friendly services layered on top.

In that sense, Amazon is not just selling processors. It is selling a platform.

How Trainium fits into Amazon’s broader AI push

Trainium is part of a larger effort by Amazon to reduce dependence on outside AI hardware and create more control over costs. The company has invested heavily in custom chips across multiple generations, and those chips are increasingly central to how AWS positions itself against rivals like Google Cloud and Microsoft Azure.

The timing is also notable. Amazon has been under pressure to show that it can keep pace in the AI race, especially as customers demand faster, cheaper access to large language models and inference services. Building its own silicon gives the company a way to lower operating costs and potentially pass some savings along to customers.

At the same time, Amazon has been broadening the model layer it offers through AWS. The source notes that Amazon formally added OpenAI to the models it serves up only recently. That suggests AWS is trying to remain platform-agnostic at the application layer while becoming more opinionated and strategic at the infrastructure layer.

The business logic of custom silicon

Custom chips can help cloud providers in three main ways:

1. They reduce reliance on the market leader, which can be expensive and supply constrained.
2. They improve margins by lowering the cost of serving high-volume workloads.
3. They create lock-in, since customers who optimize for a specific platform are less likely to leave.

Amazon’s challenge is balancing those benefits against the risk of stretching into too many businesses at once. A direct chip-selling operation would require sales support, customer onboarding, supply forecasting, and a level of hardware market discipline that cloud providers are not always built to handle.

What the move says about the AI economy in 2026

Amazon’s discussions reflect a broader reality: the AI boom is no longer just about models. It is about the physical and commercial infrastructure needed to support them. Chips, racks, data centers, power, and networking have all become strategic assets, and the companies that control them have an outsized influence on the direction of the market.

That has created a world in which cloud providers increasingly resemble semiconductor companies, and semiconductor companies increasingly resemble platform businesses. Nvidia, for its part, has not stood still. Jensen Huang has argued that the market opportunity for AI is expanding beyond GPUs alone and now includes CPUs designed specifically for AI systems. In other words, the major players are all trying to move into adjacent territory before someone else claims it first.

Amazon’s possible Trainium sales fit squarely into that pattern. The company is not just trying to buy its way out of dependence on Nvidia. It is trying to make its own hardware part of the AI market structure.

What happens next

There is still a meaningful gap between exploring an idea and shipping a product at scale. Amazon has not formally launched Trainium as an external business line, and the company says talks are still early. But the public comments from both DeSantis and Jassy suggest the concept is no longer hypothetical.

Over the next year, the key questions will be whether Amazon can expand Trainium output, whether third-party buyers show enough interest to justify a separate sales effort, and whether the economics make sense once AWS’s broader service revenue is taken into account.

If Amazon moves ahead, the real test will be whether it can convince customers that Trainium is not simply a cheaper substitute for Nvidia, but a better long-term bet for certain workloads. That would require performance, availability, software support, and trust — all at once.

For now, Amazon has sent a clear signal: it does not intend to remain a passive consumer in the AI chip market. It wants to become a seller, a platform, and possibly one of Nvidia’s most serious cloud-era challengers.

Key facts at a glance

Why this story matters for buyers, cloud customers, and investors

For enterprise customers, Amazon’s move could eventually mean more negotiating leverage in AI infrastructure procurement. For investors, it signals that the economics of cloud and chips are becoming even more intertwined. And for Nvidia, it is another reminder that its strongest competition may not come only from traditional chip rivals, but from the very cloud companies that rely on its products today.

If Amazon can scale Trainium externally without damaging AWS’s own business, it would validate a strategy that could be copied in other corners of the market. If it cannot, the company may still gain something valuable: a stronger bargaining position, deeper chip expertise, and more control over the future of its AI stack.

Either way, Amazon has made its intentions clearer. The next phase of AI competition may be fought not just in the cloud, but in the marketplace for the chips that power it.

superintelligencenews@gmail.com