Cerebras pairs its wafer-scale chips with GPUs to squeeze 5x more tokens from every megawatt
Cerebras Disaggregated Inference

Cerebras has put disaggregated inference into production: AMD Instinct GPUs handle prompt processing (prefill) while Cerebras CS-3 systems generate responses (decode). In a measured deployment on the Cerebras Inference Cloud, the split delivered 5x the throughput per megawatt of a CS-3-only setup with no loss in token speed. The company plans to roll this out across hundreds of megawatts by the end of 2027, and offers partners a full software stack or a decode-only SDK.
Cerebras takes this to wafer scale, with tens of gigabytes of SRAM and tens of petabytes per second of memory bandwidth on a single chip—350× the SRAM capacity and 5,400× the memory bandwidth of a typical GB300 GPU.