Super Micro Computer ships Vera Rubin racks featuring 1,152 GPUs

Supermicro just made a significant move in the AI infrastructure race, announcing shipments of its Data Center Building Block Solutions built around NVIDIA’s Vera Rubin platform. The flagship NVL4 configuration packs 1,152 Rubin GPUs and 576 Vera CPUs into eight liquid-cooled racks, all within a 3.2 MW power envelope.

What Supermicro actually built

The NVL4 DCBBS, announced on June 22, 2026, is built around NVIDIA’s Vera Rubin architecture, which succeeds the Blackwell generation. The Rubin GPUs bring HBM4 memory to the table, while the Vera CPUs are Arm-based chips with high core counts designed to feed data to the GPUs fast enough to keep them from sitting idle.

Supermicro’s DLC-2 direct liquid cooling technology is the engineering backbone here. Each rack handles up to 362 kW of thermal load, a figure that would overwhelm conventional air cooling entirely.

The NVL4 is not the only blueprint Supermicro has released. Earlier in 2026, the company announced DCBBS configurations around the NVL72 architecture, which puts 72 GPUs and 36 Vera CPUs inside a single rack. The NVL72 and the HGX NVL8 were both highlighted at Computex and ISC 2026. Vera Rubin products across the board are expected to reach full production volume in the second half of 2026.

Scale is the selling point

A single NVL4 deployment starts at 5 MW, but the architecture is designed to stack all the way up to gigawatt-class clusters. Supermicro’s pitch is that its DCBBS blueprints function as a standardized, pre-validated starting point, reducing the engineering work required to go from a power contract to a functioning cluster.

To support the anticipated demand, Supermicro has expanded its manufacturing footprint with new facilities in both the US and Taiwan, specifically oriented toward liquid-cooled system production.

Market reaction and what it signals

Investors responded to the announcements with notable enthusiasm. SMCI shares moved up approximately 15 to 17% following the Vera Rubin news.

Supermicro’s position in this market is worth understanding clearly. The company does not design the GPUs or CPUs at the center of these systems. NVIDIA owns that layer. What Supermicro builds is the rack-level integration: the thermal management, the power delivery, the chassis engineering, and the systems software that turns individual chips into a coherent, shippable product.

The company faced accounting and compliance scrutiny in prior years that rattled investor confidence. The current product momentum, combined with the manufacturing expansion, suggests the company is working to reestablish itself as a tier-one vendor in the AI infrastructure buildout.

The broader competitive landscape includes Dell, HPE, and a range of original design manufacturers from Taiwan. All of them are racing to qualify their own Vera Rubin-based systems.