AMD has announced Helion, an open rack compute solution designed specifically for agentic artificial intelligence workloads. The architecture allows data centers to combine CPU and GPU resources flexibly, with an all CPU configuration scaling up to 32768 cores using 6th Gen EPYC processors. Hardware availability is planned for late 2026 through ecosystem partners.
Helion builds on Open Compute Project design specifications and continues the open platform approach established by AMD Helios systems. Enterprise operators can tailor their balance of processors and accelerators to match specific agent workflows. AMD plans to submit the complete Helion design specifications directly to the Open Compute Project to allow community adoption and third party hardware development.
AMD addresses high density processing demands by packing 6th Gen EPYC chips across the chassis structure. The pure processor layout reaches 32768 cores within a single rack deployment, targeting massive agent model calculations that rely heavily on general purpose compute. AMD is currently manufacturing and validating the hardware with platform partners ahead of the planned release in late 2026.
