Hewlett Packard Enterprise (HPE) has announced a major step forward in large-scale artificial intelligence infrastructure with the launch of the industry’s first AMD “Helios” AI rack-scale architecture featuring standards-based Ethernet scale-up networking. Unveiled at HPE Discover Barcelona 2025, the solution is designed to help cloud service providers, including neoclouds, deploy high-performance AI training and inference environments faster, more efficiently, and with greater flexibility.
The new rack-scale platform combines AMD’s latest AI compute technologies with purpose-built HPE Juniper Networking hardware and software, powered by Broadcom’s Tomahawk 6 networking silicon. Built on the open Ultra Accelerator Link over Ethernet (UALoE) standard, the architecture supports trillion-parameter model training, high-throughput inference, and massive AI workloads, while avoiding proprietary interconnects that limit choice.
“For more than a decade, HPE and AMD have pushed the boundaries of supercomputing by championing open standards that accelerate innovation,” said Antonio Neri, President and CEO, HPE.
At the core of the solution is the AMD “Helios” rack-scale system, based on Open Compute Project (OCP) specifications optimized for power delivery, advanced direct liquid cooling, and high serviceability. Each rack connects up to 72 AMD Instinct™ MI455X GPUs, delivering as much as 2.9 AI exaflops of FP4 performance, 260 terabytes per second of aggregated scale-up bandwidth, and 31 terabytes of high-bandwidth memory, enabling uncompromised performance for the most demanding AI and high-performance computing workloads.
A key differentiator is HPE’s industry-first Ethernet-based scale-up switch and software, developed in collaboration with Broadcom. Designed specifically for AI workloads, the solution delivers optimized performance over standard Ethernet while reducing vendor lock-in. HPE’s AI-native automation and assurance capabilities further simplify network operations, shorten deployment timelines, and lower overall costs, complementing its broader “networks for AI” portfolio.
“With ‘Helios’, we are bringing together the full stack of AMD compute technologies and HPE’s system innovation to deliver an open, rack-scale AI platform,” said Dr. Lisa Su, Chair and CEO, AMD. “This collaboration drives new levels of efficiency, scalability, and performance for customers in the AI era.”
“Together with HPE and AMD, we are enabling powerful AI data centers built on open, Ethernet-based networking,” said Hock E. Tan, President and CEO, Broadcom.
HPE expects to make the AMD “Helios” AI rack-scale solution available globally in 2026, supported by its extensive expertise in exascale systems and liquid-cooled infrastructure deployments.