AMD used the Open Compute Project to preview Helios, a rack-scale platform built to power the next wave of AI infrastructure. The showcase wasn’t a full spec reveal, but it offered a clear look at AMD’s design philosophy: open standards, modular building blocks, and high-density performance to challenge entrenched solutions in the data center, including NVIDIA’s Rubin platform.
Helios is built on the Open Rack Wide (ORW) specification introduced by Meta, signaling a strong commitment to interoperable, vendor-neutral design. AMD’s approach blends next-generation EPYC “Venice” CPUs with Instinct MI400 AI accelerators, tied together by open fabrics and robust networking. The company emphasized an open stack, pairing UALink for scale-up connectivity with UEC Ethernet for scale-out deployments, while leaning on AMD Pensando technology to orchestrate and secure large clusters.
Cooling and serviceability are front and center. Helios employs quick-disconnect liquid cooling to sustain high thermal loads and enable dense compute while simplifying field service. The rack shown at OCP featured a double-wide ORW layout with a central equipment bay and service corridors along the sides. Horizontal compute sleds occupy an estimated 70–80% of the rack space, and the cable routing—color-coded for clarity—highlights attention to airflow, maintenance, and future upgrades.
Key takeaways
– Rack-scale platform targeting next-generation AI workloads and data center efficiency
– Built on Open Rack Wide (ORW) for open, interoperable deployment
– Combines EPYC “Venice” CPUs with Instinct MI400 AI accelerators
– Open fabrics approach: UALink for scale-up and UEC Ethernet for scale-out
– AMD Pensando for networking, security, and cluster-scale management
– Quick-disconnect liquid cooling for high density and streamlined service
– Double-wide layout with horizontal compute sleds maximizing usable capacity
While AMD kept deeper specifications under wraps, the OCP showing underscores a strategy to turn open standards into deployable systems. By unifying CPUs, AI accelerators, open fabrics, and advanced cooling in a single rack-scale blueprint, Helios aims to deliver flexibility and performance for AI training and inference at scale. With the company ramping its rack-scale solutions following its Advancing AI 2025 updates, Helios signals a serious push to compete head-on in the AI data center—where openness, density, and serviceability are quickly becoming the winning formula.
