AWS and Cerebras Collaboration Aims to Set a New Standard for AI Inference Speed and Performance in the Cloud

Amazon Web Services, Inc. (AWS), an Amazon.com, Inc. company (NASDAQ: AMZN), and Cerebras Systems today announced a collaboration that will, in the coming months, deliver the fastest AI inference solutions available for generative AI applications and LLM workloads. The solution, to be deployed on Amazon Bedrock in AWS data centers, combines AWS Trainium-powered servers, Cerebras CS-3 systems, and Elastic Fabric Adapter (EFA) networking. Later this year, AWS will also offer leading open-source LLMs and Amazon Nova using Cerebras hardware.

This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260313406341/en/

Amazon is deploying Cerebras Wafer Scale Engines in AWS datacenters​. Ultra fast inference will be available through AWS Bedrock, bringing industry leading performance to the largest hyperscale cloud.​

“Inference is where AI delivers real value to customers, but speed remains a critical bottleneck for demanding workloads like real-time coding assistance and interactive applications,” said David Brown, Vice President, Compute & ML Services, AWS. “What we're building with Cerebras solves that: by splitting the inference workload across Trainium and CS-3, and connecting them with Amazon’s Elastic Fabric Adapter, each system does what it's best at. The result will be inference that's an order of magnitude faster and higher performance than what's available today."

“Partnering with AWS to build a disaggregated inference solution will bring the fastest inference to a global customer base,” said Andrew Feldman, Founder and CEO of Cerebras Systems. “Every enterprise around the world will be able to benefit from blisteringly fast inference within their existing AWS environment.”

How It Works: Inference Disaggregation

The Trainium + CS-3 solution enables “inference disaggregation,” a technique which separates AI inference into two stages: prompt processing, or “prefill,” and output generation, or “decode.” These two stages have profoundly different computational characteristics. Prefill is natively parallel, computationally intensive, and requires moderate memory bandwidth. Decode, on the other hand, is inherently serial, computationally light, and memory bandwidth intensive. Decode typically represents the majority of inference time in these scenarios because each output token must be generated sequentially.