Drooid Logo
Back to story perspectives

Full Breakdown

Amazon Eyes Direct Sales of Trainium AI Chips, Targeting Nvidia's Dominance

6/19/2026, 6:27:57 AM

Amazon's Plan to Sell Trainium Externally

Amazon Web Services (AWS) is in early-stage talks to sell its custom Trainium AI accelerators to third-party data centers. The idea, first hinted in CEO Andy Jassy’s April shareholder letter, would shift Trainium from an internal cloud service to a standalone hardware product.

Background & Context

Trainium was introduced in 2020 as a lower-cost alternative to Nvidia GPUs for training large AI models. Customers such as OpenAI, Anthropic PBC and Uber Technologies currently access the chips through AWS. The move mirrors Alphabet’s plan to offer Tensor Processing Units (TPUs) to select external users and reflects a broader push for “sovereign” AI infrastructure after the rise of ChatGPT.

Key Figures & Groups

  • Andy Jassy – Amazon CEO, author of the shareholder letter outlining the sales concept.
  • Peter DeSantis – AWS AI chief, leading the external-sales talks.
  • Doron Aronson – AWS spokesperson, confirming the possibility of chip sales.
  • Marvell Technology – Partner in a five-year supply agreement for custom AI products.
  • TSMC – Primary foundry for Trainium, also serving Apple.
  • Nvidia Corp. – Current market leader in AI accelerators, led by CEO Jensen Huang.
  • OpenAI, Anthropic, Uber – Notable Trainium users via AWS.

Data & Statistics

  • Projected $50 billion annual run rate if Trainium were sold as a standalone product (Jassy).
  • $225 billion in revenue commitments for Trainium reported in April.
  • Amazon’s custom-silicon business, including Trainium, Graviton and Nitro, reached a $20 billion annual revenue run rate in Q1 2026.
  • Nvidia’s data-center revenue rose 92 % YoY to $75.2 billion in its fiscal Q1 2027.
  • Trainium 3 is “largely sold out,” with strong interest in Trainium 4 slated for launch next year.
  • Amazon shares rose 1.8 % to $241.82 after the news.

Official Statements & Responses

Amazon executives stress that external sales would not cannibalize AWS services, citing excess demand for AI compute. The company highlights Trainium’s cost advantage over comparable GPUs and points to a multi-year partnership with Marvell to broaden supply. Nvidia’s leadership, while not issuing a direct comment, has signaled confidence in a $200 billion AI-CPU market, suggesting parallel expansion rather than direct confrontation.

Criticism & Opposition

Analysts caution that Amazon’s entry may pressure Nvidia’s pricing but is unlikely to erode Nvidia’s dominant market share in the near term. The projected $50 billion run rate remains speculative, and Nvidia’s record data-center revenue indicates continued growth. Some view the threat as overstated given Amazon’s reliance on Nvidia GPUs for its own workloads.

Conflicting Reports & Gaps

Sources differ on the timeline for external sales; no public schedule is provided. Potential buyers are unnamed, leaving the market scope uncertain. Projections of a $50 billion run rate contrast with the $225 billion revenue-commitment figure, reflecting different measurement bases (annual sales versus total commitments).

Verbatim Quotes

  • “If our chips business was a standalone business, and sold chips produced this year to AWS and other third parties (as other leading chips companies do), our annual run rate would be ~$50 billion.” — Andy Jassy, CEO, Amazon
  • “While we’ve historically declined requests to sell chips directly, Andy noted it’s quite possible we’ll sell racks of them to third parties in the future.” — Doron Aronson, AWS spokesperson
  • “And we’re constantly looking at ways to get to more customers.” — Peter DeSantis, AI chief, Amazon
  • “There’s so much underconsumption in AI,” — Peter DeSantis, AI chief, Amazon

Why It Matters / Impact

If Amazon succeeds, enterprises will gain a cost-competitive alternative to Nvidia GPUs, potentially diversifying supply chains and reducing reliance on a single vendor. Lower-cost hardware could accelerate AI adoption across regions seeking “sovereign” compute solutions, while intensifying price competition in the accelerator market.

What’s Next

Trainium 4 is expected to debut in 2027, and Amazon plans to evaluate external sales over the next “couple of years.” Market observers will watch for announced pilot customers and any formal pricing structures as the chip-sale strategy moves from discussion to implementation.