AMD’s Helios Rack Platform Features Massive Capabilities Onboard, Rivaling NVIDIA’s Rubin Lineup
AMD made a bold statement at OCP Global Summit 2025, where it showcased its Helios rack-scale platform, a next-generation modular compute system designed to challenge NVIDIA’s Rubin lineup in the high-performance AI infrastructure market. First teased at the Advancing AI 2025 event, Helios represents AMD’s ambitious leap into rack-scale AI computing, aiming to deliver massive scalability and open-standards interoperability for data centers worldwide.
As highlighted in AMD’s official OCP blog post, the Helios platform is built on Meta’s Open Rack Wide (ORW) specification, ensuring broad compatibility and modularity across the growing open compute ecosystem. While AMD refrained from revealing exact performance specifications or configuration details, its presentation made clear that Helios is designed to rival NVIDIA’s Rubin rack system in both scale and efficiency.
“With Helios, we’re turning open standards into real, deployable systems—combining AMD Instinct GPUs, EPYC CPUs, and open fabrics to give the industry a flexible, high-performance platform built for the next generation of AI workloads.”
— Forrest Norrod, EVP and GM, AMD Data Center Solutions Group
Helios integrates AMD’s upcoming EPYC “Venice” CPUs and Instinct MI400 AI accelerators, both designed to push the limits of compute density and energy efficiency. For networking, AMD is leveraging its Pensando technology to provide scale-out capabilities, with UALink handling scale-up communication and UEC Ethernet managing scale-out workloads. This combination aims to deliver an open, vendor-neutral framework that allows large-scale data centers to build AI infrastructure without being tied to proprietary networking stacks.
AMD also confirmed that Helios will feature quick-disconnect liquid cooling, enabling higher density configurations and simplifying maintenance. This design makes it possible to service or swap components without draining the entire cooling loop, reducing downtime in high-load environments.
Images shared by AMD show that Helios uses a double-wide ORW rack layout, featuring a central equipment bay and service access areas along the sides. Inside, the rack employs horizontal compute sleds, an architecture choice differing from NVIDIA’s Kyber layout. These compute sleds occupy around 70–80% of total rack space, optimizing airflow and serviceability. Two prominent fiber-type cabling runs, color-coded Aqua and Yellow, were visible in the showcase, with distinct routing paths for power and data interconnects, reflecting AMD’s meticulous attention to cable management and modularity.
Until now, NVIDIA has largely dominated the rack-scale AI market with its Rubin and Kyber platforms. AMD’s Helios changes that landscape by introducing a truly open, flexible solution that combines CPU, GPU, and DPU integration within a single modular ecosystem. The move underscores AMD’s strategy to compete head-on in large-scale AI infrastructure, targeting hyperscalers, enterprise clouds, and research facilities seeking alternatives to NVIDIA’s closed systems.
The Helios platform reveal at OCP may have been static, but it signals an exciting new era of competition in data center architecture. As AMD and NVIDIA continue to expand their rack-scale ecosystems, the AI hardware market is set to experience rapid innovation, with open standards and modular designs leading the charge.
Do you think AMD’s Helios platform can challenge NVIDIA’s dominance in rack-scale AI systems, or will Rubin maintain the upper hand? Share your thoughts below.
