AMD Unveils Helion Agentic AI Rack for AI Agent Workloads
AMD's Helion Agentic AI Rack packs up to 32,768 Zen 6 CPU cores and 74 petaflops of FP4 performance, with availability set for November 2026.
AMD has announced the Helion Agentic AI Rack, a rack-scale computing system designed to run AI agent workloads. The company says the platform is scheduled for availability in November 2026.
The rack supports up to 32,768 Zen 6 x86 CPU cores and 16 Instinct MI350P accelerators, delivering up to 74 petaflops of FP4 computing performance and 102 Tbps of networking bandwidth. It is built on ORv3 44OU hardware compliant with Open Compute Project specifications and includes an integrated token router that directs agent model requests across frontier and open-source models.
AMD splits agent workloads into three categories. Personal-agent configurations use 256-core Venice processors at 400W, enterprise configurations use 256-core Venice processors rated at 600W, and technical computing configurations use 96-core high-frequency Venice processors.
AMD has not disclosed complete facility upgrade costs or cooling requirements for the platform.
Quick answers
When will the AMD Helion Agentic AI Rack be available?
AMD says the Helion Agentic AI Rack is scheduled for availability in November 2026.
How many CPU cores and accelerators does the Helion Agentic AI Rack support?
It supports up to 32,768 Zen 6 x86 CPU cores and 16 Instinct MI350P accelerators.
What performance does the AMD Helion Agentic AI Rack deliver?
AMD rates the rack at up to 74 petaflops of FP4 computing performance with 102 Tbps of networking bandwidth.