AWS Partners with Cerebras for AI Inference Chips

Amazon Web Services announced a multiyear partnership with AI chip startup Cerebras Systems on Friday to integrate Cerebras' Wafer-Scale Engine processors into AWS data centers, aiming to accelerate artificial intelligence inference tasks for applications like chatbots and coding tools.

The collaboration, disclosed on March 13, 2026, combines Cerebras' specialized hardware with AWS's own Trainium chips to optimize the inference process, where trained AI models generate responses to user prompts. Under the setup, Trainium processors will handle the prefill phase, converting user inputs into tokens, while Cerebras' chips manage the decode phase, producing output token by token. This division is intended to reduce latency and boost overall performance for generative AI workloads. The new capability will roll out through Amazon Bedrock, AWS's managed service for building and deploying AI applications, allowing developers to access the infrastructure without managing hardware themselves. It will support both open-source large language models and Amazon's proprietary systems, such as its Nova family, enabling uses from conversational agents to document analysis and software tools.