Amazon is exploring selling its custom Trainium AI chips to third-party data centres, according to Bloomberg and TechCrunch. AWS AI chief Peter DeSantis told Bloomberg that AWS is in talks to allow other organisations to use Trainium outside of AWS, adding that "there's so much underconsumption in AI" that external sales will not hurt cloud revenue (TechCrunch). The discussions are described as early-stage, per TechCrunch. Amazon's broader custom-silicon business -- Trainium, Graviton, and Nitro -- crossed a $20 billion annual revenue run rate in Q1 2026 at triple-digit growth, while demand for Trainium 3 is reportedly largely sold out (Bloomberg; TechCrunch). Anthropic has signed on for up to 5 gigawatts and OpenAI for around 2 gigawatts of Trainium capacity through AWS, per Bloomberg. CEO Andy Jassy wrote in his April shareholder letter that a standalone chip business selling externally could reach an annual run rate of around $50 billion (TechCrunch). Reporting frames the move as a more direct challenge to Nvidia, which currently dominates AI accelerator sales.
What happened
Amazon is exploring external sales of its custom AI chip family, Trainium, for use in other companies' data centres, according to reporting by Bloomberg and TechCrunch. AWS AI chief Peter DeSantis told Bloomberg that AWS is in talks to allow third parties to use Trainium in their own data centres; TechCrunch describes the discussions as early-stage. DeSantis told Bloomberg there is "so much underconsumption in AI" that external chip sales would not hurt Amazon's cloud business, per TechCrunch.
Current scale
Amazon's broader custom-silicon business -- covering Trainium, Graviton, and Nitro chips -- crossed a $20 billion annual revenue run rate in Q1 2026 and is growing at a triple-digit pace, per Bloomberg and TechCrunch. Demand for Trainium 3 chips is reportedly largely sold out, while customer interest for the fourth-generation chips is already building, per Bloomberg. Anthropic has signed on for up to 5 gigawatts of current and future Trainium capacity; OpenAI has agreed to around 2 gigawatts of Trainium capacity through AWS, per Bloomberg.
Revenue potential
Amazon CEO Andy Jassy wrote in his April shareholder letter that a standalone chip business selling to AWS and external customers could reach an annual run rate of around $50 billion, a figure cited in public coverage by TechCrunch. That figure is a public claim from Amazon's own shareholder communication, not an independent analyst forecast.
Competitive context
Public reporting frames the proposal as a potential competitive move against Nvidia, which currently dominates AI accelerator sales, per Bloomberg and TechCrunch. Industry-pattern observations: vendors selling custom accelerators outside their own cloud must secure high-volume manufacturing capacity from foundry partners such as TSMC, provide firmware and driver support, and deliver system-level integration guidance for rack deployments. Industry-pattern observations: cloud providers externalising accelerators must balance allocation between retail cloud customers and external purchasers, which can create lead times and capacity constraints for compute-intensive buyers.
What to watch
- •Production scale: announcements naming foundry partners or capacity-expansion deals would signal feasibility of large-scale external sales.
- •Sales packaging: whether Amazon offers standalone chips, full servers, or rack systems affects total cost of ownership and competitive positioning versus vendors selling integrated systems.
- •Customer contracts: named pilots or deployments would move the story from exploratory to operational.
For practitioners
If realised at scale, third-party Trainium sales add a cloud-tested accelerator option to the procurement landscape. Adopters should expect multi-quarter lead times and integration work typical of new accelerator families, and should monitor allocation dynamics given current Trainium 3 sell-out conditions.
Scoring Rationale #
Amazon's early-stage Trainium external sales discussions are backed by concrete demand signals: a $20 billion current custom-silicon run rate growing at triple-digit pace, plus multi-gigawatt capacity commitments from Anthropic and OpenAI already inside AWS. If the move scales, it reshapes accelerator procurement options for AI infrastructure teams and increases competitive pressure on Nvidia. Still early-stage, but the underlying fundamentals are firm and the strategic implications for practitioners are direct.
Practice with real Retail & eCommerce data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Retail & eCommerce problems