Hewlett Packard Enterprise (HPE) has unveiled the AMD “Helios” AI rack-scale solution, a comprehensive rack designed for AI training and large-scale inferencing. This innovative AI rack will feature 72 AMD Instinct MI455X GPUs, 260 terabytes of bandwidth, and an impressive 2.9 AI exaflops of FP4 performance. The solution is set to be available globally in 2026, positioning HPE as a key player in the evolving AI infrastructure market.
The collaboration between HPE and AMD spans over a decade, with both companies having previously delivered multiple exascale systems for supercomputing. Antonio Neri, CEO of HPE, highlighted the significance of open standards in this joint venture. He stated, “With the introduction of the new AMD ‘Helios’ and our purpose-built HPE scale-up networking solution, we are providing our cloud service provider customers with faster deployments, greater flexibility, and reduced risk in how they scale AI computing in their businesses.”
The AMD Helios architecture integrates hardware, software, and network capacity into a single system, ensuring a seamless experience for users. It is based on specifications from the Open Compute Project (OCP), utilizing an Open Rack Wide (ORW) design that combines energy-efficient liquid cooling with a practical maintenance setup. HPE Services will deliver the system equipped with direct liquid cooling, leveraging its experience in exascale installations.
A significant highlight of the announcement is the introduction of the HPE Juniper Networking scale-up switch, described as the first switch capable of providing optimized performance for AI workloads over standard Ethernet. Developed in collaboration with Broadcom, the switch incorporates its Tomahawk 6 networking chip. This solution is built upon the open Ultra Accelerator Link over Ethernet (UALoE) standard, allowing for the exchange of large amounts of traffic, such as high inferencing throughput and massive model sizes.
By adopting Ethernet, HPE aims to minimize vendor lock-in and facilitate faster feature updates, addressing a critical need in the AI sector. Hock Tan, CEO of Broadcom, emphasized the benefits of open Ethernet, stating, “Our high-performance silicon delivers industry-leading ultra-low latency, massive performance, and lossless networking with the scalability and efficiency modern AI workloads require.” The partnership with HPE and AMD is poised to help customers construct advanced AI data centers while maximizing choice and flexibility.
This new scale-up solution will complement HPE’s existing scale-out and scale-across offerings, creating a comprehensive “networks for AI” portfolio. The switch features AI-native automation, which is anticipated to simplify network operations significantly, enhancing overall efficiency.
The AMD Helios architecture leverages open-source AMD ROCm software alongside AMD Pensando networking technology. This approach is expected to accelerate innovation while reducing the total cost of ownership. The system is tailored to support modern AI and high-performance computing workloads, ensuring capacity and data processing capabilities remain uncompromised.
HPE primarily targets cloud service providers, including neoclouds, with this new AI solution. The demand for AI computing capacity is surging, prompting organizations to seek flexible, interoperable, and energy-efficient solutions. By combining HPE’s system innovation with AMD’s complete compute stack, the new offering aims to meet this growing demand effectively.
The global rollout of the AMD Helios AI rack-scale solution is slated for 2026, with HPE confirming it as one of the first companies to deliver such a comprehensive turnkey rack. As the landscape of AI computing continues to evolve, solutions like the Helios AI rack may play a crucial role in shaping the future of AI infrastructure.
See also
Mixx Technologies Raises $33M Series A to Revolutionize AI Infrastructure Connectivity
Apple Appoints Amar Subramanya as VP of AI, Succeeding John Giannandrea
Young Talent Fuels China’s AI Chip Revolution with 100,000 New Experts Annually
NVIDIA Invests $2B in Synopsys to Accelerate AI-Driven Engineering Innovations
Electrokare Launches AI-Driven Sports Tech for Real-Time Fatigue Detection and Recovery Insights
