AWS adopts Qualcomm chips for AI inference
Qualcomm is developing custom chips for AWS across several product generations, with AI inference as the main focus. AWS plans to add the designs to its existing Trainium, Graviton and Nitro chip families. The companies are also working on optical interconnects with bandwidth of up to 1.6 terabits, intended to manage rising data traffic in AI infrastructure.
The partnership combines Qualcomm’s experience in power-efficient chip design with Amazon’s cloud infrastructure. Qualcomm is using AWS services, including Amazon Bedrock, to accelerate its own chip-development process. For AWS, the agreement is aimed at inference workloads, where energy costs per token affect operating economics.
The deal is described as Amazon’s third major data-center win for Qualcomm since June. Meta is taking the Dragonfly C1000 server processor, while Microsoft is deploying Qualcomm’s HBC memory architecture in Azure. Qualcomm has set a goal of reaching $15 billion in data-center revenue by 2029. In March, the company’s AI research division released a framework designed to run reasoning models on smartphones.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.