On November 18, 2026, Nscale announced that its first NVIDIA Vera Rubin NVL72 systems were online, using Dell PowerEdge XE9812 servers. The announcement describes a rack-scale AI system and a managed rollout, while directing interested customers to contact Nscale about orders and timing. Nscale’s announcement

Rack-scale AI is a facility decision as well as a computing one: Luxor Technology’s general Vera Rubin estimate is 190–230 kW per rack.

How Nscale describes the rack configuration

Nscale says each rack combines 72 NVIDIA Rubin GPUs with 36 NVIDIA Vera CPUs. NVIDIA NVLink connects components within a rack, while Ethernet provides networking between racks. The systems use Dell PowerEdge XE9812 servers.

Dell separately said its integrated, liquid-cooled networking rack at Nscale was live. Dell’s statement

How Nscale says it manages the systems

Nscale says its Fleet Operations platform handles system setup, burn-in testing, repairs, retirement and real-time fleet monitoring. The company says customers can view capacity and maintenance status through the platform.

Nscale expects its global rollout to continue over the coming months. For customer inquiries, it invites interested parties to contact the company to discuss orders or timing.

The facility demands associated with Vera Rubin

Luxor Technology explains Vera Rubin rack architecture and general data-center power and cooling requirements.

Luxor Technology’s general Vera Rubin overview estimates power demand at 190–230 kW per rack and describes direct-to-chip liquid cooling as required. These are platform-level estimates and guidance, not measurements of Nscale’s installed racks.