Dell has introduced a new high-end AI server built around Nvidia’s Vera Rubin NVL4 architecture, positioning it as a key component of its broader enterprise AI platform. The PowerEdge XE8812 is designed for customers planning large-scale AI infrastructure deployments and can scale to 144 GPUs per rack.code.visualstudio

The server will sit at the center of the Dell AI Factory with Nvidia, a preconfigured stack that combines servers, storage, networking, and software into a single package. That stack typically includes Dell PowerEdge AI servers, Nvidia GPUs such as H100, H200, Blackwell, and now Vera Rubin-class systems, plus Ethernet or InfiniBand networking, Dell PowerScale and PowerStore storage, and software including Nvidia AI Enterprise and NIM inference microservices.code.visualstudio

More Memory, More Density

Dell says the liquid-cooled XE8812 delivers a major jump in compute density and memory capacity. Compared with the previous GB200 NVL4 generation, the Vera Rubin platform adds more host memory, more cores, more GPU memory, and more compute. Dell says the new design increases cores from 144 to 176 and provides 50% more memory per socket and GPU memory.code.visualstudio

That extra memory is important because it allows organizations to run larger AI models and simulations entirely in memory. Dell says that reduces the need for staging or swapping data between host memory and storage, which can add latency and cut effective bandwidth in AI and HPC workloads.code.visualstudio

Built for AI and HPC Convergence

Dell is also emphasizing the overlap between AI and high-performance computing. The company says those workloads are growing beyond what incremental hardware upgrades can handle, especially as enterprises demand more data, compute, and control inside their own environments.code.visualstudio

The XE8812 includes iDRAC for remote deployment, updates, and monitoring. At the rack level, Dell offers the Integrated Rack Controller and OpenManage Enterprise, which provide telemetry and automated leak detection to help identify problems early.code.visualstudio

Dell cites Gartner data showing AI investment is expected to rise sharply in 2026, with spending on AI-optimized servers projected to increase 49%, while AI infrastructure as a whole could add $401 billion in spending.code.visualstudio

Nvidia’s Vera Rubin Platform Expands

Dell’s announcement is part of a broader rollout of Nvidia’s Vera Rubin architecture. Nvidia says the platform combines compute, networking, and data processing into rack-scale systems designed for large AI data centers. The stack includes the Vera CPU, Rubin GPU, NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch, and Groq 3 LPU, all integrated into a system intended to function like an AI supercomputer.code.visualstudio

Nvidia says the architecture is meant to support the full AI lifecycle, from large-scale training and post-training to real-time inference. It is aimed at “AI factory” deployments and other large-scale data center builds.code.visualstudio

Super Micro Joins the Rollout

Super Micro also used the Vera Rubin launch to announce its own AI server plans. The company says its liquid-cooled design will support up to 1,152 Nvidia Rubin GPUs and 576 Nvidia Vera CPUs in a rack-scale configuration. That system will be part of Super Micro’s Data Center Building Block Solutions (DCBBS) Blueprint, which covers compute, networking, liquid cooling, power distribution, and site planning guidance for AI infrastructure deployments.code.visualstudio

Super Micro says its blueprint is based on experience building large liquid-cooled projects quickly, including on-site facility surveys that evaluate loading docks, data hall dimensions, floor load ratings, and available power and cooling capacity.code.visualstudio