Amazon shares rise as report says AWS exec is pushing Trainium sales to outside data centers

Amazon’s stock rose by nearly 3% in the session after a report claimed AWS executive Peter DeSantis is targeting NVIDIA customers as Amazon pushes to expand the reach of its own AI chip strategy. The post circulated via Yahoo Finance, which attributed the idea to discussions around how Amazon could sell its custom silicon beyond Amazon’s own cloud services.

At the center of the claim is Trainium, Amazon’s purpose-built artificial intelligence accelerator. In simple terms, accelerators are specialized processors that handle the heavy math of machine learning training, and AI workloads increasingly depend on them for speed and cost efficiency. Trainium is already used inside Amazon Web Services for training jobs running on AWS infrastructure, but the report suggests a broader distribution model.

According to the report’s summary, Amazon is discussing selling Trainium chips directly to external corporate data centers. That would be a structural shift from a pure “use our cloud” model toward a “buy our hardware” option for companies that prefer to run AI systems on premises or in their own managed facilities.

The same report framed the effort as an attempt to reach customers currently using NVIDIA hardware. While the post did not provide details on which specific NVIDIA users were being targeted, the implication is that Amazon wants to insert Trainium into the procurement pipeline of enterprises with existing investment in NVIDIA-based AI stacks.

For NVIDIA, the threat would not necessarily be an immediate loss of AWS training demand, but the potential to expand the set of platforms where Trainium could be considered. NVIDIA’s advantage in this category is closely tied to its broad software and hardware ecosystem, which customers often treat as a long-term standard when they plan multi-year AI infrastructure programs.

Industry context: as generative AI adoption grows, enterprises are weighing three common infrastructure approaches. They can run on hyperscalers like AWS, buy proprietary or specialized accelerators and deploy on premises, or mix vendors using interoperable software stacks. A credible pathway for Trainium to reach outside data centers would place Amazon more directly in the “procure accelerators” conversation that often happens long before workloads are even scheduled.

Still, major specifics were not included in the reported account. The post summary did not outline expected timelines, pricing, supply arrangements, or whether Amazon would offer turnkey hardware-software integration, support services, or references for adoption. It also did not disclose how Amazon would address compatibility and performance for customers that already standardize on NVIDIA platforms.

What to watch next is whether Amazon confirms the discussions, provides a product or commercial framework for external sales, or indicates how it will handle deployment and support. Investors will likely look for evidence that any plan can scale beyond pilots, because the hardest part of switching AI infrastructure is typically not the hardware purchase itself, but integration with existing tooling, orchestration, and long-running training pipelines.