AMD showcases Helios, a fully integrated rack-scale AI infrastructure
AMD Helios is a robust double-width rack-scale system for AI workloads. Helios was recently shown at the Supermicro Computex booth.
AMD Helios is the company’s new and fully integrated rack-scale AI infrastructure. The solution was announced at Supermicro’s booth in the recent Computex 2026 show in Taiwan.
AMD Helios is optimized for large-scale AI deployments. To that end, it delivers high levels of compute density and performance for frontier model training and high-throughput inferencing.
AMD Helios will be a component offering of Supermicro’s Data Center Building Blocks Solution (DCBBS), according to a recent presentation by Supermicro CEO Charles Liang.
What’s Under the Helios Hood?
AMD describes Helios as a 72-GPU double-width rack-scale system. It’s powered by AMD Instinct MI455X GPUs (forthcoming soon), 6th Gen AMD EPYC CPUs (also coming soon), and AMD Pensando networking technologies.
In Helios, they’re unified by the AMD ROCm open software stack.
Key capabilities of Helios include:
- Modular scalability from rack to cluster level
- Open networking for both scale-up and scale-out AI
- Advanced security
- Integrated virtualization with rack-scale software acceleration
At rack scale, a Helios system will deliver up to 1.4 exaFLOPS of FP8 and 2.9 exaFLOPS of FP4 performance, AMD says. Helios also offers 31 TB of total HBM4 memory and 1.4 PB/s of aggregate bandwidth.
AMD calls these specs a “generational leap” that enables trillion-parameter training and large-scale AI inferencing.
Which Standards Underlie Helios?
Helios is built fully on the Open Rack Wide (ORW) form factor standard.
ORW—originally introduced by Meta and then specified by the OCP—is designed for AI-scale data centers. It defines an open, double-wide rack optimized for the power, cooling and serviceability demands of next-gen AI systems.
Helios is the first rack-scale design from AMD engineered specifically for frontier AI workloads. To that end, AMD says, Helios provides both hyperscalers and enterprises a future-proof, open-standard based platform that unites power, flexibility and interoperability.
AMD Helios is designed to be more than just a hardware reference; it’s also a collaboration blueprint for the AI ecosystem.
Helios enables OEM and ODM partners to:
- Adopt and extend the Helios reference design, accelerating time-to-market for new AI systems
- Integrate AMD Instinct GPUs, EPYC CPUs and Pensando DPUs with their own differentiated solutions
- Participate in an open, standards-based ecosystem that drives interoperability, scalability and long-term innovation
Who are Helios Partners?
Among the companies making early announcements to partner is Tata Consultancy Services of India. Earlier this year, AMD announced it would work with Tata’s HyperVault subsidiary to co-develop a rack-scale AI infrastructure design based on Helios for India’s national AI initiatives.
Another important partner for AMD Helios is Supermicro. The company is one of the first partners collaborating closely with AMD to bring the Helios solution to market. It’s another example of Supermicro’s leadership in delivering end-to-end AI infrastructure solutions.
Supermicro showcased AMD Helios in its booth at the recent Computex show in Taiwan. And CEO Liang mentioned Helios several times during his recent Supermicro Innovate keynote address.
AMD is also working with other Helios partners, especially in Taiwan. Last month AMD said it would invest more than $10 million to support manufacturing in Taiwan for Helios. The investment will go toward expanding partnerships and scaling advanced packaging manufacturing, AMD says.
What are New Hardware Components in Helios?
AMD Helios will be powered by two AMD components that are forthcoming: the AMD Instinct MI455X GPU and the 6th Gen AMD EPYC CPU family, codenamed Venice. Though neither has been formally announced yet, AMD has discussed both publicly for months.
For example, last month AMD said it has begun production ramp for the 6th Gen EPYC CPU using TSMC’s 2nm process technology in Taiwan. In the future, AMD added, it will expand production of the new EPYC CPU family at TSMC’s fab in Arizona.
AMD also said Venice is the first high-performance computing product to use the fab’s 2nm process tech. Industry sources expect Venice to be formally announced in July and then ship later in the year.
AMD has been a bit more forthcoming around the Instinct MI455X GPU. The company says key features will include:
- Architecture based on the latest 4th Gen AMD CDNA
- Memory using Samsung's HBM4, providing up to 13 Gbps processing speed and 3.3 TB/s bandwidth
- Performance expected to deliver substantial improvements in throughput, memory efficiency and compatibility.
- Applications will be ideal for next-generation AI systems, combining AMD Instinct GPUs, AMD EPYC CPUs, and rack-scale architectures—including the AMD Helios platform
DO MORE:
- Learn more about Supermicro and AMD Helios
- Read an AMD Helios blog: Advancing openness in AI infrastructure