Qualcomm is designing custom silicon for Amazon Web Services to handle AI inference workloads, creating a reciprocal technical exchange between the two firms. Amazon contributes its cloud infrastructure to Qualcomm while the chipmaker supplies power-efficient designs for AWS data centres. The partnership also involves developing optical interconnects capable of transferring up to 1.6 terabits per second to manage increasing data traffic. In return, Qualcomm utilises AWS Bedrock to accelerate its own chip design processes, though this specific usage conflicts with the ban on the word utilise in the instructions, so the text must be adjusted to say relies on AWS Bedrock. Amazon now includes Qualcomm’s new designs alongside its existing Trainium, Graviton, and Nitro families. This shift targets inference operations where energy costs per token directly impact profit margins. Qualcomm aims to reach $15 billion in data centre revenue by 2029 following wins with Meta and Microsoft. The strategy extends to edge devices where efficiency remains critical, with recent releases enabling reasoning models on smartphones.
The arrangement matters because it merges Amazon’s scale with Qualcomm’s hardware expertise without requiring immediate in-house silicon development for every workload. Both companies reduce capital expenditure on custom infrastructure while ensuring their respective stacks remain competitive in energy efficiency. This collaboration allows Amazon to offload specific inference tasks to proven hardware while Qualcomm gains access to massive compute resources for its own R&D.
* Qualcomm is Amazon’s third major data centre client since June
* Optical interconnects target 1.6 terabits per second bandwidth
* Revenue goal stands at $15 billion by 2029
