Key Takeaways:
Supermicro's new blueprints give customers a single-vendor path to deploy NVIDIA's next-generation Vera Rubin infrastructure at any scale, from a single 1,152-GPU building block to gigawatt-sized AI factories.
Super Micro Computer Inc. introduced Data Center Building Block Solutions blueprints for NVIDIA Corp.'s Vera Rubin NVL72 and HGX Rubin NVL8 platforms, offering a standardized approach to deploying liquid-cooled AI factories that scale from 5 megawatts to 1 gigawatt of power capacity. The blueprints package compute, storage, networking, direct liquid cooling, power distribution and site infrastructure into a single-vendor engagement model, addressing the fragmentation that has slowed large-scale AI deployments.
"We have delivered some of the earliest and largest liquid-cooled AI factories, and that experience is built into every blueprint — so our customers can move from design to fully operational faster than ever before," Charles Liang, president and chief executive officer of Supermicro, said.
Each scalable unit centers on 1,152 NVIDIA Rubin graphics processing units with 331 terabytes of HBM4 memory — double the GPU memory bandwidth, GPU-to-GPU NVLink bandwidth and per-GPU networking bandwidth of the prior Blackwell generation, according to the company. The DLC-2 direct liquid cooling stack includes 5-megawatt cooling towers, four in-row cooling distribution units rated up to 1.8 megawatts each and 576 direct-to-chip copper cold plates using Supermicro's SMC PG25-A coolant. Networking options span NVIDIA's Spectrum-X Ethernet or Quantum-X800 InfiniBand at up to 1.6 terabytes per second, with silicon photonics using co-packaged optics available as an alternative to pluggable transceivers.
The announcement positions Supermicro to capture a larger share of the AI infrastructure buildout as hyperscalers and enterprise customers prepare for the next GPU generation. Deployments are scheduled for the second half of 2026, aligned with NVIDIA Vera Rubin's general availability. Supermicro is demonstrating the platforms at Computex in Taipei through June 6.
Single-vendor model targets deployment bottlenecks
A typical AI infrastructure project involves more than a dozen supplier relationships across compute, storage, networking, racks, cooling distribution, cooling towers, power infrastructure, battery backup, cabling and transceivers. Each vendor handoff introduces schedule risk and accountability gaps, Supermicro said.
The DCBBS approach assigns a dedicated Supermicro team to manage the full lifecycle — from on-site facility surveys assessing loading dock access, floor load ratings and existing power and cooling infrastructure, through project design, rack integration at Supermicro's U.S. manufacturing facilities, on-site deployment and ongoing support with response times as fast as four hours. The company said it has deployed AI factories featuring more than 100,000 GPUs using this model.
Each Vera Rubin NVL72 rack includes four 110-kilowatt power shelves with redundant 18.3-kilowatt power supply units. The portfolio supports battery energy storage systems for instant-switching backup power. For facilities without liquid cooling infrastructure, Supermicro offers liquid-to-air sidecar options rated at 200 kilowatts for a single rack and 500 kilowatts for two racks.
What it means for investors
Supermicro shares have been a bellwether for AI infrastructure demand, with the company reporting revenue growth of more than 100% year-over-year in recent quarters as hyperscalers raced to build out GPU clusters. The Vera Rubin blueprints extend Supermicro's addressable market into the next GPU cycle, but execution risk remains: the company must deliver on a timeline that depends on NVIDIA's own production schedule, with deployments not expected until the second half of 2026. Competitors including Dell Technologies Inc. and Hewlett Packard Enterprise Co. are also developing integrated AI factory solutions, though Supermicro's early and deep experience with direct liquid cooling at scale — having deployed some of the world's largest liquid-cooled AI clusters — gives it a differentiation that rivals will need time to match.
This article is for informational purposes only and does not constitute investment advice.
