AMD and Cerebras Systems have announced a technical partnership pairing AMD's Helios rackscale system with Cerebras' Wafer-Scale Engine (WSE) in a disaggregated inference solution. The combined configuration delivered up to five times the tokens per second per watt (TPS/W) in internal testing, addressing a Cerebras WSE efficiency challenge by offloading prompt processing to AMD's offering. The efficiency claims are compared against an existing Cerebras WSE baseline while running the open-source Kimi 2.6 1T model, released on 20 April 2026, a mixture-of-experts design with one trillion total parameters but only 32 billion active per token, shipping natively in INT4. At INT4, the full weight set runs to roughly 500 GB, while a single Cerebras wafer holds 44 GB, meaning a Cerebras-only deployment needs somewhere north of a dozen wafers just to hold the model, whereas one Helios rack could hold it around sixty times over. The partnership gives AMD access to SRAM decode technology without spending the $20 billion Nvidia shelled out at the end of last year for a non-exclusive deal. The article notes the absence of any financial information regarding the partnership.