Supermicro expands OCP-based rack-scale portfolio for faster AI data center rollout
Super Micro Computer, Inc. has announced an expanded portfolio of Open Compute Project Accepted systems, racks and rear door heat exchangers (RDHx), backed by its Data Center Building Block Solutions (DCBBS). According to the company, the offering is meant to shorten the time it takes to bring AI and HPC infrastructure online.
- Supermicro expanded its OCP Accepted portfolio of systems, racks and rear door heat exchangers under DCBBS, available now.
- ORv3 and ORW racks use a centralized 48V DC busbar and DLC-2 Direct-to-Chip liquid cooling for higher density.
- RDHx cooling spans 10 kW to 120 kW, and a 250 kW 4U in-rack CDU is part of the range.
- Networking includes 800G and 1.6T Ethernet switches, Broadcom Tomahawk 6 CPO platforms and NVIDIA Spectrum-X Ethernet Photonics.
- The 2 OU NVIDIA HGX B300 design fits up to 18 nodes and 144 GPUs per rack.
What the portfolio covers
The announcement, dated 14 October 2026 from San Jose, California, describes an integrated range built on the Open Compute Project (OCP). It includes rack-scale AI and HPC systems, ORv3 and ORW racks, 48V DC power, networking and liquid cooling. Supermicro says the pieces can be combined into end-to-end deployments at rack and cluster scale, and that the OCP solutions are available now.
The ORv3 and ORW racks replace per-server power supplies and cables with a centralized 48V DC busbar. According to Supermicro, this lets customers tune their CPU and GPU systems for higher-density racks. The company pairs this with its DLC-2 Direct-to-Chip liquid cooling, which it says removes heat more efficiently from CPUs, GPUs and other critical components.
Charles Liang, president and CEO of Supermicro, said that by integrating compute, networking, power, advanced cooling, software and services into validated rack-scale solutions, the company helps customers "simplify complexity" and speed the path from innovation to production. Dirk Van Slyke, vice president of Marketing at OCP, said customers need open, flexible infrastructure that can grow with the changing demands of AI.
Cooling, networking and management
Supermicro's RDHx units mount on the back of ORv3 and ORW racks and remove residual heat from the server exhaust before it enters the data hall. The company says this works without extra changes to the data hall's air conditioning. The cooling portfolio spans 10 kW to 120 kW. DC models connect directly to the rack busbar to reduce cabling, and the designs include N+1 redundant fans and a smart anti-condensation system.
On networking, the DCBBS portfolio includes air- and liquid-cooled 800G and 1.6T Ethernet switches. These include next-generation Co-Packaged Optics (CPO) platforms built on Broadcom Tomahawk 6 switching silicon, optimized for EIA or ORv3 racks. The Supermicro Enterprise Advanced SONiC Networking Operating System lets customers choose the network architecture that fits their needs.
DCBBS will also include NVIDIA Spectrum-X Ethernet Photonics switch systems, described as the first co-packaged Ethernet switches with 200 Gb/s SerDes. They are built on the 102.4 Tb/s NVIDIA Spectrum-6 Ethernet ASIC. According to the release, they offer 5x better energy efficiency, 10x longer MTBI, 5x longer uptime for AI applications and 1.3x faster deployment time compared with networks based on pluggable transceivers. Spectrum-X Ethernet Photonics is now in production.
For management, operators can monitor and manage the full set of DCBBS components through SuperCloud Composer using industry-standard protocols. Software including SuperCloud Director, SuperCloud Automation Assistant and SuperCloud Developer Console is positioned for managing AI factories, enterprise data centers and NeoCloud infrastructure.
Selected products
- 1 OU FlexTwin: a rack-scale HPC solution with 1 OU 2-node systems, powered by Intel Xeon 6 or AMD EPYC CPUs.
- NVIDIA Vera Rubin NVL72: the NVIDIA MGX 1.2 Rack series, a Supermicro contribution to the OCP program, designed for NVIDIA Vera Rubin NVL72 and NVIDIA Vera CPU platforms.
- 2 OU NVIDIA HGX B300: compact 8-GPU nodes built to the OCP ORv3 specification with liquid cooling, fitting 21-inch racks and allowing up to 18 nodes and 144 GPUs per rack.
- 2 OU 4-node rack-scale solution: aimed at 24/7 agentic AI, with up to 45,696 Arm AGI CPUs per rack.
- 250 kW 4U In-Rack CDU: cooling of up to 250 kW in a 4U form factor, supporting 45 °C warm-water cooling.
- OCP ORv3 Rack series: 44OU and 48OU 21-inch racks with integrated busbars and a certified static load capacity of 5,000 lb (2,268 kg).
- AMD Helios platform solution: a fully integrated ORW architecture for large-scale inference, frontier model training, fine-tuning, and sovereign and agentic AI.
The range also includes ORv3 and ORW chiller doors, with 20-100 kW and 60 kW cooling capacity respectively, plus two ready-made PG coolants for the TCS loop: DLC-1 for standard liquid-cooled racks and the ultra-low-conductivity DLC-2 for dedicated cold-plate designs. Both are designed to meet the OCP PG25 base specification.
What it means for operators
Supermicro's pitch is speed and simplicity: one validated, open foundation covering systems, busbars, and rack- and data-center-level cooling. According to the company, DCBBS lets customers shorten Time to Market (TTM) and Time to Operate (TTO). The 10 kW to 120 kW RDHx range is presented as a practical way to raise rack density in both new AI facilities and existing data centers. How these claims hold up in practice will depend on individual deployments, as the release provides no pricing or customer results.
