Amazon ropes Qualcomm into something, something AI, networking chips

Amazon has enlisted mobile chip pioneer Qualcomm to co-develop specialized chips for artificial intelligence inference and high-speed optical networking gear. The agreement could result in payments totaling up to $60 billion over the next nine years. Qualcomm's CEO, Cristiano Amon, expressed excitement about collaborating with AWS to bring advanced processing and power-efficient compute solutions to the market.

Although the specifics of the partnership remain undisclosed, Qualcomm's involvement could entail the creation of custom silicon, chiplets, or tailored versions of its datacenter accelerator portfolio. The new rack-scale AI platform aims to replace expensive high-bandwidth memory (HBM) with high-bandwidth compute, enabling higher effective memory bandwidth for AI inference workloads using more affordable LPDDR memory.

This technology is set to launch next year as part of Qualcomm's AI250-series of Dragonfly rack systems. Additionally, Qualcomm is developing a new datacenter CPU, the C1000, which is expected to reach 5 GHz and have over 250 cores by the second half of 2028. In terms of networking, AWS will utilize Qualcomm's serializer-deserializers (SerDes) and optical digital signal processors (DSPs) to increase port speeds to 1.6 Tbps.

This collaboration highlights the growing importance of high-speed optical interconnects for scaling AI infrastructure. Furthermore, Qualcomm plans to leverage Amazon's Bedrock platform for electronic design automation workloads, contributing to the circular economy within the AI industry.