Amazon weighs selling external access to its Trainium AI chips, Bloomberg says, lifting attention on Marvell

Amazon is exploring whether it could sell its custom-built Trainium AI chips to customers outside of its own cloud business, according to a Bloomberg report carried by Yahoo Finance on June 29. The item says the discussions are active, placing additional focus on semiconductor suppliers that market data-center AI infrastructure.

The report frames Amazon’s Trainium effort as part of a broader strategy to control key components of its artificial intelligence stack. Trainium is Amazon Web Services’ (AWS) purpose-built processor design for machine learning workloads, intended to reduce latency and improve performance for AI training and inference in the cloud.

Bloomberg’s report, as summarized by Yahoo Finance, indicates Amazon is now considering external sales of those chips rather than limiting Trainium strictly to internal use in AWS. The exact structure of any offering was not detailed in the Yahoo Finance item, and Amazon did not publicly describe terms, timelines, or customer targets in the material available for this story.

The market-news framing also points to Marvell Technology as a company investors are watching as the chip landscape evolves. Marvell is known for supplying data-center semiconductor products that support networking and compute, including components used in AI systems. However, the Yahoo Finance summary does not state that Marvell would be directly involved in any Trainium sales plan, only that it is drawing attention amid the report.

For AWS, selling access to custom silicon could be a way to extend its AI platform beyond software. If Amazon offers chips or chip-enabled capacity to enterprises and cloud customers, it could potentially create a new revenue stream tied to AI infrastructure, while also deepening vendor relationships around its AI services.

At the same time, custom chips are often tightly coupled to the operating environment around them, including how they integrate with memory, networking, and software tooling. The Yahoo Finance report does not indicate whether Amazon would sell bare silicon, provide chip-embedded systems, or include services that package the processors into a broader platform. Without those specifics, it is unclear how broadly Trainium could fit into existing customer stacks built around other suppliers.

Amazon did not provide additional disclosure in the Yahoo Finance item beyond the claim that discussions are ongoing, and no details were included about pricing, manufacturing partners, or the competitive positioning of Trainium relative to alternatives. Investors and customers will likely look for clearer indicates from Amazon, such as announcements about product format, availability, and how AWS would support the chips for external buyers.

What to watch next is whether Amazon confirms the concept and, if so, clarifies the sales model. Key items include whether Trainium would be offered as standalone processors or bundled with systems and software, and whether customers would gain choice in AI hardware without being locked into a single cloud environment.