Dell teams up with NVIDIA to launch “full liquid cooling” AI servers! Featuring the Vera Rubin architecture debuting on the system/rack for the first time, with 144 GPUs per single rack—computing power beyond capacity

Dell Technologies announced a deepening partnership with NVIDIA at the ISC Conference held in Hamburg, Germany, launching the PowerEdge XE8812 server equipped with the new Vera Rubin NVL4 architecture. The product features a 100% liquid cooling, fanless design, with a single rack capable of housing up to 144 GPUs, specifically designed to meet the integrated demands of next-generation AI and HPC high-performance computing, with global release expected in early 2027.
(Background: OpenAI negotiating a 20-year ultra-long lease to lock in prices, NVIDIA betting on Ohio 10GW data center with a "lease first, pay later" model)
(Additional background: Jensen Huang declined to testify before Congress, with U.S. lawmakers sarcastically saying: "He has time to eat fried sauce noodles, but no time to explain NVIDIA’s business in China")

Table of Contents

Toggle

• First to adopt Vera Rubin architecture, all-liquid cooling challenges performance limits
• Favored by the U.S. Department of Energy, over 5,000 institutions worldwide adopt
• Executives optimistic about the big integration of AI and HPC

As global AI investment is projected to grow at an astonishing 44% annually by 2026, traditional data center infrastructure is gradually struggling to handle the massive and complex computational demands. To address this, tech giant Dell Technologies (NYSE: DELL) officially unveiled its latest addition to the Dell AI Factory lineup — the Dell PowerEdge XE8812 server — at the ISC Conference in Hamburg, Germany, on June 22, 2026. This hardware, designed specifically for the most demanding high-performance computing (HPC) and AI workloads, marks the beginning of a new era in global supercomputing infrastructure.

First to adopt Vera Rubin architecture, all-liquid cooling challenges performance limits

The core highlight of the PowerEdge XE8812 is its adoption of NVIDIA’s latest Vera Rubin NVL4 architecture. Compared to the previous generation, it not only increases the core count from 144 to 176, but also boosts memory capacity per slot and per GPU by 50%. This means large language models and complex scientific simulations can run entirely in memory, eliminating bottlenecks caused by data staging or swapping, greatly reducing system latency.

In terms of cooling and space configuration, the server uses 100% direct liquid cooling and a fanless design, supporting over 300kW of astonishing power, providing extremely high energy efficiency. By deploying Dell PowerRack 9100 cabinets based on OCP open standards, a single cabinet can house up to 144 GPUs, maximizing computing power while minimizing space. Coupled with pre-integrated validated deployment services from the manufacturer, deployment can be completed and operational in as little as 6 hours.

Favored by the U.S. Department of Energy, over 5,000 institutions worldwide adopt

Dell and NVIDIA’s strong partnership has already gained significant commercial momentum. Currently, more than 5,000 enterprises and institutions worldwide have deployed Dell AI Factory. Among the most notable cases is the U.S. Department of Energy’s (DOE) next-generation flagship supercomputer “Doudna” at Lawrence Berkeley National Laboratory. This system is based on the XE8812 and Vera Rubin NVL4 architecture, combined with NVIDIA Quantum-X800 InfiniBand network, aiming to accelerate scientific breakthroughs from molecular biology to astronomy.

Not only in the U.S., but this powerful computing infrastructure has also expanded globally. It includes France’s InstaDeep with the Kyber supercomputer for AI model training and automated design (performance reaching 0.5 exaFLOPs), the UK’s Wellcome Sanger Institute’s large-scale genome decoding project, and Australia’s Monash University’s MAVERIC supercomputer used for cancer and climate research, all heavily reliant on Dell’s new solution.

Executives optimistic about the big integration of AI and HPC

Regarding this revolutionary hardware upgrade, Arun Narayanan, Senior Vice President of Dell’s Computing and Networking Business, emphasized that institutions conducting the world’s most important research need infrastructure that matches their ambitions, and the XE8812 can help them accomplish seemingly impossible tasks. Chris Marriott, Vice President of NVIDIA’s Enterprise Platform, added that the integration of AI and HPC is redefining market expectations for hardware, and the alliance between the two will provide the industry’s most efficient and high-performance open ecosystem.

According to an official announcement, this highly anticipated Dell PowerEdge XE8812 server is expected to be officially launched globally in early 2027, which will surely add fuel to the global AI computing race.