Qualcomm said it has entered a multi-generational partnership with Amazon to develop customized chips for large-scale AI data centers, alongside optical interconnect technology aimed at advancing AI inference workloads.

According to Qualcomm, the partnership is centered on custom chips for large-scale AI data centers and is designed to support AI inference, an area that has become increasingly important as companies look for more efficient infrastructure to run production AI applications.

The companies also plan to jointly develop optical interconnect solutions with capacity of up to 1.6T and future generations. That places connectivity alongside processor design in the scope of the agreement, suggesting the collaboration is not limited to compute silicon alone.

Qualcomm said it will use AWS AI infrastructure, including Amazon Bedrock, for electronic design automation workloads as part of an effort to shorten chip design cycles. In the company’s announcement, Chief Executive Cristiano Amon said rising AI demand requires breakthroughs in both computing and connectivity. AWS Vice President Prasad Kalyanaraman said the partnership is intended to provide customers with more efficient and cost-effective infrastructure.

The announcement did not include financial terms, a product launch schedule, customer availability details, or deployment targets. It also did not specify whether the customized chips are intended primarily for Amazon’s internal infrastructure, external cloud customers, or both.

Why It Matters

The deal adds to the broader shift toward specialized AI infrastructure, where chip design, networking, and cloud platforms are being developed in closer alignment. For Amazon, the agreement could deepen its options in AI infrastructure buildout. For Qualcomm, it marks a higher-profile push into data center AI systems, an area where performance, power efficiency, and system-level integration are increasingly central.

The partnership also matters because it highlights that AI infrastructure competition is extending beyond model development into the hardware stack that supports inference at scale. Custom silicon and interconnect design are becoming more important as cloud providers seek tighter control over cost, workload optimization, and supply strategy.