AMD and Cerebras announce disaggregated inference solution

AMD and Cerebras will pair AMD Helios systems with the Cerebras Wafer-Scale Engine in one disaggregated inference workflow. The companies claim up to 5x higher tokens per second per watt. The joint solution is expected to reach Cerebras Cloud customers in the second half of 2026, so it is a product-timing story more than a near-term revenue event.

• Combines AMD Helios with the Cerebras Wafer-Scale Engine
• Claimed up to 5x higher tokens per second per watt
• Initial availability through Cerebras Cloud in 2H 2026