Amazon AWS to Deploy Cerebras Wafer-Level AI Chip

Amazon AWS and AI chip company Cerebras Systems recently announced a collaboration to deploy the Cerebras CS-3 wafer-level AI system within AWS data centers. This system will work in conjunction with Cerebras' proprietary Trainium chip to provide AI inference services through the Amazon Bedrock platform.

The solution employs an "inference deconstruction" technique: Trainium handles prompt processing, while Cerebras CS-3 generates the output. Both are connected via AWS's high-speed network. The CS-3 is equipped with the WSE-3 chip, boasting 900,000 AI cores and 21PB/s memory bandwidth. The service is expected to launch in the second half of 2026, providing customers with Cerebras hardware-accelerated versions of mainstream open-source large models and the Amazon Nova model.