Revolutionizing Rack-Scale Computing

AMD is set to transform data center infrastructure with the introduction of its Helion Agentic AI Rack. This modular computing system is specifically engineered to power the next generation of artificial intelligence agents. By offering highly customizable configurations, the platform aims to address the varying demands of personal, enterprise, and technical computing environments.


The system boasts impressive technical specifications, supporting up to 32,768 Zen 6 x86 CPU cores and 16 Instinct MI350P accelerators. This hardware is capable of delivering 74 petaflops of FP4 computing performance, supported by a networking bandwidth of 102 Tbps, all housed within an ORv3 44OU hardware framework that adheres to Open Compute Project standards.


Tailored Solutions for Diverse AI Workloads

AMD has categorized agent-based workloads into three distinct tiers, each requiring specific hardware optimization:

  • Personal Agents: Focus on high density and power efficiency using 256-core Venice processors at 400W.
  • Enterprise Applications: Balance capacity and performance using 256-core Venice processors at 600W.
  • Technical Computing: Prioritize raw processing speed via 96-core, high-frequency Venice processors.

To manage this complexity, the enterprise-grade configuration utilizes distinct server nodes to separate agent execution, inference via MI350P accelerators, and general-purpose tasks such as database and storage management. Furthermore, the architecture includes an integrated token router to efficiently distribute requests across various frontier and open-source models.


Infrastructure Challenges for Modern Data Centers

The introduction of such high-density hardware poses significant hurdles for conventional data centers. Most existing facilities were designed for traditional cloud or legacy enterprise computing and may struggle with the following:

«Traditional data centers may lack the necessary power delivery, cooling capacity, and network infrastructure to support racks that integrate tens of thousands of CPU cores alongside intensive GPU configurations,» as noted by industry observers.

Beyond the thermal and power requirements, the diversity of the hardware—mixing high-core-count processors with high-frequency systems—forces data center operators to move away from uniform deployments. Instead, they must carefully plan the allocation of power, cooling, and space for mixed server types.


Additionally, the massive 102 Tbps throughput and the requirement for a programmable token router necessitate upgrades to networking hardware and software integration. While the Helion Agentic AI Rack is slated for release in November 2026, AMD has yet to provide comprehensive details regarding the specific costs of facility upgrades or definitive cooling mandates.