AMD Unveils Helios: Next-Gen Rack-Scale AI Platform Challenges NVIDIA's Dominance

AMD has made a significant move in the AI hardware space with the unveiling of its Helios rack-scale platform at the Open Compute Project (OCP) Global Summit 2025 in San Jose, California. This new offering is positioned to directly compete with NVIDIA's rack-based GPU and CPU combinations, such as the Blackwell and Grace-powered GB300 NVL72 1

1

Source: DT

Helios Platform Specifications

The Helios platform combines AMD's latest hardware technologies, including:

• AMD Epyc server CPUs
• Instinct MI450 GPUs based on CDNA architecture
• AMD Pensando networking interfaces
• AMD ROCm software

Each MI450 GPU boasts up to 432GB of HBM4 memory with a total bandwidth of 19.6 TB/s. A fully equipped Helios system, containing 72 GPUs, is capable of delivering:

• 1.4 exaFLOPS of FP8 performance
• 2.9 exaFLOPS of FP4 precision
• 31TB of total HBM4 memory
• 1.4PB/s of aggregate bandwidth
• Up to 260TB/s of internal interconnect throughput
• 43TB/s of Ethernet-based scaling capacity 3

  3

AMD claims this configuration offers 50% greater memory capacity than NVIDIA's next-generation Vera Rubin system and up to 17.9 times higher performance than its previous generation 4

4

Open Rack Wide Specification and Design

Helios is built on Meta's newly introduced Open Rack Wide (ORW) standard, which aims to improve power efficiency, cooling, and serviceability for AI systems. The design features:

• A double-wide layout for reduced weight density and improved serviceability
• Liquid cooling with quick-disconnect capability for sustained thermal performance
• Standards-based Ethernet scale-out for multipath resiliency and interoperability
• Compliance with OCP DC-MHS, UALink, and Ultra Ethernet Consortium frameworks 3

  3

Source: TechRadar

Market Impact and Partnerships

The introduction of Helios represents a significant challenge to NVIDIA's dominance in the AI hardware market. AMD has already secured a major partnership with Oracle, which plans to deploy 50,000 MI450 GPUs starting in Q3 2026, with more to follow in 2027 1

1

Forrest Norrod, Executive VP and GM of the Data Center Solutions Group at AMD, emphasized the importance of open collaboration in scaling AI efficiently. He stated, "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" 5

5

Source: Wccftech

Related Stories

HPE adopts AMD's Helios rack architecture with custom Ethernet switch for 2026 AI systems

AMD Helios AI rack system takes direct aim at Nvidia with Microsoft, OpenAI backing

AMD unveils Helios rack-scale AI system with 72 MI455X accelerators and 256-core EPYC Venice

AMD Challenges NVIDIA with Ambitious Rack-Scale AI Accelerators for 2026

Industry Reception

The Helios system garnered significant attention at the OCP Summit, with ServeTheHome's Patrick Kennedy noting large crowds gathering around the display. The reference design on show was valued at approximately $3 million 2

2

While AMD's Helios platform shows promise, its success will ultimately depend on broad ecosystem support and whether it can truly establish itself as an industry standard rather than just another branded interpretation of open infrastructure 3

3