Amazon may take Trainium chips beyond AWS
June 20, 2026
Amazon is reportedly discussing sales of its Trainium AI chips to external data centers. That would be more than cloud marketing: it targets Nvidia’s hardware power.
What this is about
Amazon is exploring the sale of its own Trainium AI chips to other data-center operators, according to a Bloomberg report carried by the Taipei Times on 20 June 2026. Until now, Trainium has mainly been an AWS offering: customers rent compute in Amazon’s cloud instead of running the hardware themselves.
If Amazon actually takes this step, it would not be a minor product variant. It would shift the line between cloud provider and chip supplier. Amazon would no longer only say: use our cloud. It would also say: use our AI hardware where you need direct control over the data center.
What Trainium actually does
AWS Trainium is Amazon’s own AI accelerator for training and inference. It is designed to run large models more economically and efficiently than general-purpose processors. Amazon points to Anthropic, OpenAI, Uber, Odyssey, DeCart AI, Neura Robotics, Splash Music and Poolside as users or partners in the Trainium ecosystem.
The new point is the possible sale outside AWS. Peter DeSantis, Amazon’s senior vice president for AI models and semiconductors, said talks with potential customers are under way, though he did not name them. Trainium3 is largely sold out, and Amazon says there is already strong interest in the next generation.
Why it matters
Nvidia dominates the AI accelerator market. Many companies want alternatives, but alternatives need more than a good chip. They need supply chains, software, developer tools, cooling, power planning and enough manufacturing capacity. Amazon has one advantage through AWS: it operates large infrastructure itself and learns from real workloads.
For Europe, the possible move is especially interesting. Techzine frames it as part of the demand for sovereign AI infrastructure: organizations want to run AI capacity locally or regionally without depending completely on a US hyperscaler. Ironically, Amazon could serve that demand if it sells hardware to local operators.
In plain language
Imagine a bakery that has used its special ovens only to bake its own bread. Now it is considering selling those ovens to other bakeries. It would then compete not only with other bakeries, but also with the companies that make ovens.
A practical example
A fictional European cloud provider runs 5,000 GPUs for AI customers and wants a second source of capacity. If 20 percent of new capacity could run on Trainium racks, that would be 1,000 accelerators not tied directly to Nvidia or large US clouds. Whether that works economically depends on software compatibility, utilization and electricity costs.
Scope and limits
First, only talks are known so far, not confirmed external mega-customers. Second, Amazon cannot scale without limits if advanced chip manufacturing capacity remains tight. Third, a cheaper chip does not replace a mature developer platform; many AI teams stay where their models and tools run reliably.
The sovereignty angle is also mixed. Local operators would get more physical control, but the hardware and part of the software ecosystem would still come from a US company. That is better than pure cloud dependency, but not the same as full technological independence.
SEO & GEO keywords
Amazon Trainium, AWS, AI chips, Nvidia, AI infrastructure, data centers, sovereign AI, AI accelerator, semiconductors, TSMC
💡 In plain English
Amazon has mostly used Trainium as an advantage inside its own cloud. If the chips move into other data centers, that becomes a more direct push into the AI hardware market.
Key Takeaways
- →Amazon is reportedly discussing selling Trainium chips for data centers outside AWS.
- →The move would position Amazon more directly as a hardware supplier against Nvidia.
- →Sovereign and locally controlled AI infrastructure in Europe is one possible driver.
- →Trainium3 is largely sold out, according to Amazon executive Peter DeSantis.
- →Production capacity remains the bottleneck because Amazon also depends on advanced chip manufacturing.
FAQ
Is Amazon already selling Trainium directly?
The reports describe talks and exploration. No confirmed external data-center customers were named.
Why does this matter for Europe?
If Trainium can run outside AWS, regional providers could build more local AI capacity without relying entirely on Nvidia or US clouds.
Is Trainium an Nvidia killer?
No. Nvidia remains dominant. The interesting part is that Amazon may move from cloud rental toward supplying AI hardware directly.