Supermicro, SpaceXAI Plans Point to Gigawatt-Scale AI Buildout
Supermicro’s Vera Rubin systems are designed to scale from 5 megawatts to gigawatt capacity as SpaceXAI expands infrastructure for Grok.
Supermicro is shipping NVIDIA’s next-generation Vera Rubin NVL72 systems as AI infrastructure moves toward data centers measured in gigawatts rather than megawatts.
The company’s Data Center Building Block Solutions are designed to scale from a 5-megawatt deployment to gigawatt capacity. A standard scalable unit includes 1,152 Rubin GPUs, 331 terabytes of HBM4 memory and 16 compute racks, with liquid cooling, power delivery, storage and networking integrated into the design.
The announcement comes as SpaceXAI expands its AI plans around NVIDIA’s Vera Rubin platform and moves toward gigawatts of computing capacity. It also plans to adapt an NVL72 system for its first-generation Starmind AI satellite.
Supermicro said each NVL72 rack combines 72 Rubin GPUs and 36 Vera CPUs in a liquid-cooled system. The design packages computing, networking, storage, power delivery and liquid cooling into standardized deployment units.
The scale of the plans highlights the infrastructure pressures created by advanced AI. Gigawatt-scale facilities require large power connections, extensive cooling systems and coordinated delivery of compute, networking and storage equipment.
Supermicro reported more than $60 billion in new orders during the quarter ended June 30, 2026, but that figure covers the company’s broader business and does not identify SpaceXAI’s share. Its fiscal fourth-quarter revenue was $11.1 billion, with a 17.5% gross margin.
The public disclosures therefore confirm a shared technology direction, but not a finalized Supermicro-SpaceXAI construction project. They do not identify a site, power supplier, project cost or completed capacity.
