AMD Debuts Helion Rack With 32,768 Zen 6 Cores for AI Servers

AMD has introduced Helion, a new standardized reference design engineered specifically for rack-scale agentic AI workloads. Building upon the infrastructure foundation established by the Helios design unveiled at the OCP Global Summit last year, Helion targets the escalating compute density requirements of modern data center operations. The architecture utilizes a 21-inch rack form factor and is engineered to support CPU-only configurations housing up to 32,768 Zen 6 cores. This extreme scale is achieved through the integration of the 256-core AMD EPYC 9996 processor, part of the Venice microarchitecture family. The reference design will be available in two primary configurations: a dedicated CPU-only deployment optimized for high-throughput reasoning tasks and an all-in-one system tailored for integrated workload acceleration. AMD plans to begin shipping Helion reference platforms next year, offering data center operators a standardized blueprint for next-generation AI infrastructure. By providing a clear, scalable pathway from single-node deployments to massive rack-scale architectures, Helion positions AMD to address the growing demand for agentic AI systems that require sustained, high-density computational throughput. The announcement reinforces AMD's strategy of delivering purpose-built data center hardware capable of handling increasingly complex machine learning and autonomous agent workloads as industry infrastructure requirements continue to evolve.