Supermicro Shipped Liquid-Cooled NVIDIA Rubin Racks
The company has begun deploying data center systems built on the Vera Rubin platform for high-density AI clusters.
Updated on Sept. 23, 2026 in Data Centers
Live Poll
Do you trust companies currently leading the AI hardware buildout as stable long-term investments?
Supermicro has officially commenced shipping its liquid-cooled data center racks featuring the new NVIDIA Vera Rubin hardware. This deployment marks the initial stage of integrating the Rubin platform into high-compute infrastructure.
Why it matters
The system architecture enables scalable cluster deployments from 5 MW up to gigawatt-class capacities, addressing the power and thermal demands of modern AI training. This expanded production aligns with rising industry requirements for high-performance computing density.
The NVL4 configuration consists of 1,152 Rubin GPUs and 576 Vera CPUs distributed across eight liquid-cooled racks. Each unit operates within a 3.2 MW power envelope, utilizing direct liquid cooling to manage thermal loads.
The players
Supermicro
A high-performance server manufacturer that specializes in modular, liquid-cooled rack architectures for AI and data center applications.
NVIDIA
A semiconductor company that designs the Rubin GPU and Vera CPU architectures powering current large-scale generative AI infrastructure.
The details
Supermicro employs DLC-2 (direct liquid cooling) technology, a method that uses liquid coolant circulating directly to heat-generating components to manage thermal loads efficiently. The NVL72 architecture integrates 72 GPUs and 36 Vera CPUs per rack, creating modular units that scale from a 5 MW starting deployment. To streamline integration, the company provides pre-validated system blueprints that reduce the engineering requirements for data center operators.
Timeline
June 22, 2026: Supermicro announced the NVL4 DCBBS configuration.
September 23, 2026: News coverage regarding Vera Rubin rack shipments was published.
Second half of 2026: Vera Rubin products are expected to reach full production volume.
The Tech Race
The transition to the Rubin platform represents the latest evolution in the race to optimize hardware for massive-scale AI training. It succeeds existing Blackwell-based systems by providing higher compute density and the thermal management necessary for gigawatt-class infrastructure.
The move to full-scale production in late 2026 will primarily affect large-scale data center operators and enterprise cloud providers requiring high-density clusters. Customers can leverage pre-validated blueprints to reduce custom engineering overhead during site installation.
The takeaway
The Vera Rubin platform moves from initial shipment to mass manufacturing throughout the end of 2026. Industry observers should monitor the Q4 2026 supply chain reports to confirm if production volume meets the projected demand for multi-megawatt AI clusters.
Further reading
For more on how infrastructure is evolving to meet modern compute demands, visit Data Centers.
Live Poll
Do you trust companies currently leading the AI hardware buildout as stable long-term investments?
