Amazon Web Services is taking a more direct step into the AI hardware race, with plans that could place its in-house Trainium chips in the hands of outside customers. The move would bring AWS closer to the center of a market long led by Nvidia.
According to AWS AI chief Peter DeSantis, the company is in early discussions about offering Trainium for use in other data centers. Amazon says the idea grew from strong demand for its custom chips, which CEO Andy Jassy highlighted in his annual shareholder letter.
Jassy noted that Amazon's chip business could reach an annual run rate of about $50 billion if it operated as a standalone unit serving AWS and third parties. He also said demand has been so strong that current Trainium capacity sold out quickly, along with capacity for the next-generation Trainium4.
For Amazon, the opportunity goes beyond chip sales. AWS already earns from the broader ecosystem around AI workloads, including storage, networking, security, and monitoring. That makes its hardware strategy part of a wider cloud platform play.
Still, expanding supply for outside buyers would require enough manufacturing capacity to support both AWS customers and new partners. Amazon has said it may sell racks of chips in the future, while keeping its long-term focus on scaling its own AI infrastructure.
As cloud providers deepen their hardware ambitions, the next phase of AI competition may be shaped not only by software and models, but by who controls the chips powering them.
