AI
Nscale brings first NVIDIA Vera Rubin NVL72 systems online for AI cloud workloads

Nscale has brought its first NVIDIA Vera Rubin NVL72 systems online, combining Dell servers and liquid-cooled networking as it prepares to expand next-generation AI computing capacity globally.
[For more news, click here]
Building more powerful AI systems increasingly depends on how efficiently computing hardware, networking and cooling technologies work together. Large AI models require significant processing capacity, while reasoning models and autonomous AI agents are creating additional demand for infrastructure capable of handling complex tasks.
AI cloud provider Nscale has brought online its first NVIDIA Vera Rubin NVL72 systems, marking the beginning of a planned global deployment over the coming months. The infrastructure uses Dell PowerEdge XE9812 servers and is integrated with Nscale's cloud platform to support AI model training, inference and increasingly complex reasoning applications.
The company has also become the first to deploy Dell's fully integrated, liquid-cooled networking rack, designed to simplify the installation and operation of high-performance AI infrastructure. However, Nscale has not disclosed the location of its initial deployment or the total computing capacity expected from the planned rollout.
Vera Rubin brings new computing capabilities
The NVIDIA Vera Rubin NVL72 architecture combines 72 Rubin graphics processing units (GPUs) and 36 Vera central processing units (CPUs) within a single rack-scale system. The components work together through high-speed connections, allowing the infrastructure to handle demanding AI workloads that require substantial computing power and fast data exchange.
The platform integrates NVIDIA's latest computing and networking technologies, including NVLink 6 and Spectrum-X Ethernet, to support communication between processors and across multiple racks. This is particularly important for large-scale AI training and inference, where delays in transferring information can reduce overall system performance.
According to NVIDIA, the Vera Rubin NVL72 architecture is designed to improve the efficiency of AI training and deliver higher inference throughput compared with its previous Blackwell generation. The company claims the platform can achieve up to 10 times higher inference throughput per watt under specified workloads, although those performance claims do not represent independently verified results from Nscale's deployment.
For cloud customers, the additional computing capacity could support larger AI models, advanced reasoning systems and applications that rely on AI agents performing multiple tasks. Actual performance will depend on workload requirements, software configuration and how efficiently the infrastructure is operated.
Dell introduces liquid-cooled networking infrastructure
A key part of the deployment is Dell's PowerRack for Networking, which brings computing network components, power distribution and liquid cooling into a factory-integrated system.
Built on the Dell IR9148 platform, the networking rack includes eight NVIDIA-based PowerSwitch SN6600-LD switches, alongside power management equipment, an integrated controller and an in-rack coolant distribution unit. Dell says the system provides more than 819 terabits per second of aggregate switching capacity.
Unlike traditional deployments that require networking components to be assembled and tested individually at data centers, Dell's approach allows the complete rack to be configured and validated before shipment. The company says this reduces installation complexity and helps operators bring new infrastructure into service more quickly.
Liquid cooling is also becoming increasingly important as advanced AI processors generate substantial heat and require greater computing density. Integrating cooling directly into the networking infrastructure could help operators manage these requirements, although Dell has not disclosed independently measured energy savings from Nscale's installation.
Nscale prepares for wider AI infrastructure rollout
The new systems are being incorporated into Nscale's existing cloud management environment, which monitors computing resources, network performance and infrastructure availability.
Through its Fleet Operations platform, the company manages system configuration, testing, maintenance and performance monitoring. This is intended to allow additional Vera Rubin systems to operate within a common environment as the global deployment expands.
The rollout comes as AI cloud providers invest in infrastructure designed for more demanding workloads, particularly models that process longer sequences of information and AI agents that perform increasingly complex operations.
Nscale expects to bring additional Vera Rubin systems online in the coming months, although it has not specified the number of planned installations, deployment locations or confirmed customer commitments. The company says customers can begin discussing orders and deployment schedules.
The initial installation represents an operational step in bringing NVIDIA's latest architecture into commercial AI cloud infrastructure. Its broader significance will depend on how successfully Nscale expands the platform and whether the new systems deliver measurable improvements in computing performance, operational efficiency and customer workloads.





