AI Models & Platforms

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

Theo Nash
Add Unite.AI to your preferred sources on Google
Key PointsNscale is first to deploy Dell's factory-integrated liquid-cooled L11 networking rack.
View more
  • Nscale is first to deploy Dell's factory-integrated liquid-cooled L11 networking rack.
  • NVLink handles scale-up inside each rack while RoCE Ethernet connects racks into larger clusters.
  • NVIDIA claims up to 10 times more tokens per megawatt than GB200 NVL72 for its stated configuration.

Nscale announced on November 18, 2026 that its first NVIDIA Vera Rubin NVL72 rack-scale AI systems were online, beginning a global rollout expected to continue over the coming months. The Dell PowerEdge XE9812 systems add capacity for customers training and serving large models, including workloads that reason through multistep problems.

Each NVL72 rack contains 72 Rubin GPUs and 36 Vera CPUs, together with memory and high-speed connections designed to function as one rack-scale supercomputer. A companion technical post identifies large distributed training, long-context reasoning, reinforcement learning, agentic inference, and low-latency services processing high token volumes as target workloads. That range makes the network and operating model important: the infrastructure has to coordinate many accelerators for both training runs and production inference.

Factory-Integrated Racks and Networking

Nscale says it is the first in the industry to deploy Dell’s fully integrated, liquid-cooled L11 networking rack. Built on Dell’s IR9148 platform, it combines eight NVIDIA SN6600-LD switches with shuffle boxes, power shelves, an integrated rack controller, and an in-rack coolant distribution unit. Dell integrates and tests the populated rack at the factory, eliminating onsite network-rack assembly.

“Dell PowerRack integrates and validates compute, networking, power and cooling in the factory, reducing the work required onsite and helping Nscale bring its high performance compute infrastructure into production faster,” said Ihab Tarazi, CTO of Compute and Networking in Dell Technologies’ Infrastructure Solutions Group.

The computing hardware follows the same rack-scale approach. Dell’s system overview describes the XE9812 as the Vera Rubin successor to its XE9712, using direct liquid cooling and NVLink-interconnected GPUs in a rack-scale design. In Nscale’s deployment, NVLink provides the scale-up fabric within each rack, while a RoCE scale-out network connects racks into a larger cluster.

RoCE, short for RDMA over Converged Ethernet, supplies high-throughput, low-latency communication for distributed AI traffic. Nscale’s Ethernet fabric incorporates high-performance switching, congestion management, telemetry, and automation on a standards-based foundation. The design is intended to keep GPUs working on parallel jobs rather than waiting on memory, networking, or operational bottlenecks.

Fleet Controls for the Global Rollout

The racks enter Nscale’s cloud operating model as soon as they come online. Fleet Operations manages enrollment, configuration, validation, burn-in, network integration, maintenance, repair, and retirement. Control Center applies software, firmware, network, and security baselines before validated systems enter service through an automated process.

Observability spans compute, networking, storage, and orchestration, linking infrastructure telemetry with automated responses. Nscale’s Radar API exposes availability, maintenance and repair status, device state, capacity, and other operational events to customers. Those controls give customers a common view of capacity while providing Nscale with a consistent way to bring additional systems into service.

Nidhi Chappell, Nscale’s president of AI infrastructure, said the company expects the rollout to operate as one production environment under the same fleet automation and observability as the rest of its cloud. Customers can contact Nscale to place orders or discuss deployment timing.

On NVIDIA’s product page, the vendor reports up to 10 times more tokens per megawatt than GB200 NVL72 for its stated Kimi-K2 Thinking configuration with 32K/8K input/output sequence lengths, while noting that the inference performance is subject to change. The vendor comparison—not a result measured on Nscale’s systems—puts the emphasis on token throughput within a fixed power footprint, a central economic consideration for large-scale inference.

The deployment connects a new GPU generation to the less visible work of making it usable: factory-integrated liquid-cooled networking, NVLink within racks, Ethernet scale-out, and shared fleet controls. The next operational test is whether Nscale can repeat that package as additional systems arrive, expanding customer capacity without fragmenting the global production environment.

Theo Nash is an AI-generated research agent at Unite.AI, covering AI infrastructure, compute, and the hardware systems that power modern artificial intelligence. His work focuses on the technical foundations behind large-scale AI workloads, including data centers, accelerators, networking, and the software stacks that tie them together.

With an analytical and engineering-driven perspective, Theo examines how advances in GPUs, custom silicon, memory architectures, and distributed systems enable new generations of AI models. He pays particular attention to performance trade-offs, energy efficiency, scalability, and the practical constraints that shape real-world deployment of AI infrastructure.

Articles authored by Theo Nash are AI-generated and reviewed by Unite.AI’s editorial team to ensure technical accuracy, clarity, and responsible coverage of the rapidly evolving AI compute landscape.