FinalAI-edited source brief

Amazon Is in Early Talks to Sell Its Trainium AI Chips to Outsiders, Targeting a $50 Billion Business

AWS confirmed discussions to sell its homegrown Trainium processors to external data-center operators, a strategic pivot that would turn an internal cost center into a standalone rival to Nvidia.

Published ...7 sources0 Reddit6 web85% confidence

What matters

AWS confirmed early talks to sell Trainium AI chips to outside data-center operators.
CEO Andy Jassy estimates a standalone chip business could reach a ~$50 billion annual run rate.
Trainium2 is already a multi-billion-dollar business with 1M+ chips in production; Trainium3 was unveiled in December 2025.
Amazon remains a massive Nvidia customer, with a deal for 1 million GPUs across AWS by the end of 2027.
Capacity constraints raise questions about Amazon’s ability to supply external buyers without slowing AWS growth.

What happened

On June 18, 2026, AWS AI chief Peter DeSantis told Bloomberg that Amazon is in early-stage talks to sell its Trainium AI chips to third-party companies for deployment in non-Amazon data centers, according to TechCrunch. The discussions stem from Amazon CEO Andy Jassy’s annual shareholder letter in early April, where he wrote that demand for Amazon’s homegrown silicon was strong enough that the company might begin selling racks of chips to outsiders. Jassy estimated that if Amazon’s chip unit were a standalone business selling to AWS and third parties, it would generate an annual run rate of roughly $50 billion. DeSantis declined to name potential buyers, and Amazon emphasized that the talks are preliminary.

The announcement builds on months of public momentum. At AWS re:Invent in December 2025, Amazon unveiled Trainium3, which it said is four times faster and more power-efficient than Trainium2, while Jassy disclosed that Trainium2 was already a multi-billion-dollar run-rate business with more than one million chips in production and over 100,000 companies using it through Amazon’s Bedrock platform. In his April letter, Jassy noted that AWS’s overall AI revenue had surpassed a $15 billion run rate as of Q1 2026, but that growth was being held back by capacity constraints despite adding 3.9 gigawatts of data-center capacity in 2025 with plans to double that by 2027.

Why it matters

For years, Amazon designed custom silicon primarily to lower its own infrastructure costs and reduce reliance on outside suppliers. Selling Trainium externally would transform that strategy, turning an internal efficiency play into a direct challenge to Nvidia’s dominance in AI accelerators. Jassy has explicitly drawn a parallel to Amazon’s Graviton CPUs, which since their 2018 launch have eaten into Intel’s data-center dominance; he argues the “same story arc is unfolding in AI.”

The move also reflects growing customer demand for alternatives to Nvidia’s GPUs. Jassy wrote that customers want better “price-performance,” and claimed Trainium3 is 30–40% more price-performant than its predecessor. While Nvidia currently commands a market on a roughly $326 billion revenue run rate—dwarfing Amazon’s projected chip revenue—the emergence of a credible, cloud-native challenger with Amazon’s scale could pressure pricing and procurement strategies across the industry. Notably, Amazon remains one of Nvidia’s largest customers, with a deal to take delivery of one million Nvidia GPUs by the end of 2027 even as it courts the same buyers with its own silicon.

Public reaction

No significant Reddit or public discussion threads were available for this story. Early industry coverage has focused on the strategic implications for Nvidia and the feasibility of Amazon balancing internal demand with external chip sales.

What to watch

The most pressing question is whether Amazon can manufacture enough Trainium chips to supply external buyers without starving its own cloud growth. Jassy has acknowledged that AWS is already capacity-constrained, and two large customers have reportedly asked to buy all of Amazon’s custom Graviton CPU capacity for 2026—requests Amazon had to decline. Observers should also monitor which data-center operators emerge as the first named buyers, and whether Nvidia responds with pricing or partnership adjustments as its biggest cloud customer becomes a direct competitor.

Sources

Public reaction

Open questions

Which data-center operators are in talks with AWS?
Can Amazon manufacture enough Trainium chips to serve external buyers without starving AWS growth?
Will Nvidia adjust pricing or partnership terms as Amazon becomes a direct competitor?

What to do next

Developers

Benchmark Trainium2 instances (Trn2) on AWS for inference workloads to measure cost and latency savings against GPU-based alternatives.

Understanding real-world price-performance now will let you migrate quickly if Amazon expands external chip availability.

Founders

Use Amazon’s emerging silicon strategy as leverage in pricing discussions with incumbent GPU cloud providers.

A credible second-source narrative strengthens your negotiating position even before external Trainium sales begin.

PMs

Map your AI product roadmaps against multi-cloud silicon options and treat Trainium as a plausible secondary inference target by 2027.

Diversifying chip dependencies reduces vendor lock-in and hedges against Nvidia supply constraints.

Investors

Model Amazon’s chip unit as a nascent semiconductor business with ~$50B run-rate potential, but discount heavily until firm external customers and supply commitments are announced.

The opportunity is large, but early-stage talks and capacity constraints mean revenue timing is uncertain.

Operators

Audit your current AWS spend to identify inference workloads that could migrate to Trainium2/3 instances.

Migrating compatible workloads to Amazon’s native silicon can lower compute costs today without waiting for external hardware availability.

Testing notes

Caveats

This story covers a strategic business announcement regarding future chip sales; no product, API, or general-purpose developer tool has been released for external use.
Reported talks are in early stages, and no external buyers or commercial timelines have been disclosed.