AMD has introduced Helion, a new rack-scale computing design aimed at meeting the growing infrastructure demands of agentic AI. Announced on Monday (October 12, 2026), the platform gives data center operators and enterprises greater flexibility to run AI agents that can reason, use tools, and execute multi-step tasks with limited human intervention.
Unlike conventional AI applications that primarily respond to user prompts, agentic systems can perform sequences of actions, often requiring repeated model calls and substantial computing resources. AMD Helion is designed to address these demands through configurable server infrastructure built around the company’s processors and AI acceleration technology.
Flexible Configurations for Different AI Workloads
AMD Helion is based on an open rack design aligned with Open Compute Project (OCP) specifications. Its architecture allows customers to adjust CPU core density, computing performance, and power requirements according to their workloads.
The company outlined two primary configurations. A CPU-only setup targets hyperscale deployments requiring dedicated processing capacity for large numbers of AI agents. An integrated configuration combines CPU and GPU resources for enterprise workloads that need both general-purpose computing and accelerated AI processing.
Up to 32,768 CPU Cores in One Rack
The CPU-only configuration can accommodate up to 32,768 cores using AMD’s sixth-generation EPYC processors, according to the company’s announcement. The integrated option combines EPYC CPU nodes with AMD Instinct MI350P GPU nodes and general-purpose computing resources.
Helion also supports both air-cooled and liquid-cooled deployments, giving data center operators options based on their infrastructure requirements. These configurations help organizations manage demanding AI workloads while balancing computing capacity, power consumption, and deployment needs.
Open Software and Model Routing
Another feature is an integrated token router that directs AI agents’ model requests across frontier and open-source models. This approach helps organizations coordinate model usage within agent-based workflows.
AMD’s open-platform strategy also gives hardware partners room to develop systems around a common design. Companies supporting the initiative include Cisco, Dell Technologies, HPE, Lenovo, Supermicro, QCT, Wiwynn and Hon Hai Technology Group.
Also Read: AMD to Buy Fei-Fei Li’s World Labs in USD 8.2 Billion AI Deal
Availability and Industry Outlook
AMD said it is developing Helion configurations with ecosystem partners, with availability planned for November 2026. The company also intends to contribute the specification to the Open Compute Project to encourage wider industry participation.
The announcement highlights how AI infrastructure is evolving beyond GPU-heavy systems towards architectures that combine CPUs, accelerators and software coordination. As businesses adopt increasingly autonomous AI tools, flexible rack-scale platforms could become an important part of supporting these workloads.