Under the terms of the new deal, Cerebras chips will be deployed within Amazon Web Services (AWS) data centers. These processors will be interconnected with Amazon’s proprietary Trainium3 custom AI chips using specialized networking technology developed by Amazon.
Every customer large or small is on AWS, from individual developers to the largest banks in the world, and this deal will make it easy as a click to get on Cerebras.
The joint effort focuses on optimizing AI inference, the stage where trained models process user requests to generate answers. The companies are splitting this workload into two distinct phases: a \"prefill\" stage, where user input is converted into tokens, and a \"decode\" stage, where the system generates the final output. Amazon’s Trainium3 chips will handle the prefill process, while Cerebras hardware will manage the decoding phase, a method Cerebras CEO Andrew Feldman described as a \"divide and conquer\" strategy.
This technical setup mirrors a strategy expected from Nvidia, which is anticipated to detail plans for combining its graphics processing units with technology from Groq, a startup it acquired for $17 billion in late December. While the timeline for the Nvidia-Groq integration remains uncertain, Amazon expects its Trainium3 program to begin production workloads within the coming months.
What we can say is that we believe Trainium3—and future Trainium4—will continue to lead in price-performance versus merchant GPUs.
The new service is scheduled to become available in the second half of this year. While financial specifics of the deal were not disclosed, Amazon indicated that it expects its integrated hardware solution to provide superior value compared to upcoming competitor offerings.