Amazon Web Services (AWS) has partnered with Qualcomm to build custom data center chips optimized for AI inference across multiple product generations. Under the agreement, AWS will integrate Qualcomm’s hardware alongside its existing Trainium, Graviton, and Nitro chip lines, targeting energy-efficient inference workloads. In return, Qualcomm is adopting AWS cloud services, including Amazon Bedrock, to accelerate its internal semiconductor design processes.
The partnership also includes joint development of high-speed optical interconnects supporting bandwidth up to 1.6 terabits to relieve data bottlenecks in cluster architectures. The deal marks Qualcomm’s third major hyperscaler agreement since June, following server processor commitments with Meta and cloud memory infrastructure deployments with Microsoft Azure.
Qualcomm aims to generate $15 billion in annual data center revenue by 2029 through its focus on power efficiency per token. As cloud operators face rising power costs and hardware demands, alternative chip partnerships are becoming central to managing large-scale model deployment economics.
Why it matters
Highlights hyperscaler strategies to diversify inference hardware beyond dominant providers to lower cost per token.
Demonstrates the growing trend of semiconductor manufacturers using generative cloud tools like Bedrock to design hardware.
Validates Qualcomm’s expansion into enterprise data center infrastructure targeting high-bandwidth, power-efficient workloads.
Source: the-decoder.com
