Nscale Activates Its First Vera Rubin NVL72 Racks for Global AI Cloud

Unite.AI
Read full story· 4 min
Nscale activated its first NVIDIA Vera Rubin NVL72 rack-scale AI systems on November 18, 2026, starting a global rollout of Dell PowerEdge XE9812 systems designed for large AI model training and inference workloads. These racks integrate 72 Rubin GPUs and 36 Vera CPUs with advanced networking and cooling, enabling efficient distributed AI processing.

What happened

On November 18, 2026, Nscale activated its first NVIDIA Vera Rubin NVL72 rack-scale AI systems. This marks the start of a global rollout of Dell PowerEdge XE9812 systems, which are designed for large-scale AI model training and inference. Each rack integrates Rubin GPUs and Vera CPUs with advanced networking and cooling, and the rollout is expected to continue over the coming months. The systems are immediately managed through Nscale’s cloud operating model, providing additional capacity for customers.

Key details

  • The systems use Dell PowerEdge XE9812 hardware with direct liquid cooling and NVLink interconnects.
  • Dell’s IR9148 platform provides integrated, factory-tested networking with eight NVIDIA SN6600-LD switches per rack.
  • Nscale’s Radar API gives customers access to operational data like capacity, maintenance, and device status.
  • NVIDIA claims up to 10 times more tokens per megawatt for the Rubin NVL72 compared to GB200 NVL72 in certain configurations, though this is vendor data and not measured on Nscale’s systems.

Why it matters

Nscale’s customers, including organizations training and deploying large AI models, now have access to new infrastructure that can handle more complex and larger-scale workloads. The fully integrated racks and automated fleet controls reduce setup time and operational complexity for both Nscale and its clients. This deployment increases available compute capacity and may lower operational costs for high-volume AI inference and training tasks.


More in Chips & Compute