Dell ships a liquid-cooled networking rack for NVIDIA Vera Rubin NVL72
Dell Technologies says its PowerRack for networking is now running in full production at Nscale, as part of an NVIDIA Vera Rubin NVL72 scalable unit. The company describes it as the first OEM networking rack that is fully validated and cooled directly with liquid. This is data-centre hardware, not a consumer product, but it shows where AI infrastructure is heading: the network is now sold as a finished rack, not as a pile of parts.
What is in the rack
According to Dell, the PowerRack for networking combines switches, cabling, power distribution, a PowerCooled RCDU and management software in one factory-integrated rack. Dell does not spell out the abbreviation in its post. The idea is that the customer connects water, power and fabric cabling and then deploys the rack as part of a larger AI unit, instead of sourcing, configuring and validating each element separately.
Each Dell IR9148 rack holds eight PowerSwitch SN6600-LD switches, together with shuffle units and scale-out fabric leaf and spine configurations. The switches are built on the NVIDIA Spectrum-6 ASIC, the switching chip, and are fully liquid-cooled. Dell says each ASIC delivers 102.4 Tb/s of switching capacity, which adds up to more than 819 Tb/s per rack.
Why cooling sits in the rack
As GPU systems grow, the network has to keep distributed AI workloads supplied with bandwidth. Dell argues that traditional approaches to thermal management hit limits as density rises, and that direct liquid cooling, where coolant runs to the hot components, should be designed into the rack from the start. Dell says it validates cooling, power and networking together, so that operators do not treat cooling as a separate integration job.
The post is written by Ihab Tarazi, Dell’s chief technology officer for its Infrastructure Solutions Group Core. He calls the benefits a more predictable path from delivery to production and less integration risk. Those are Dell’s claims, and the post offers no measured figures for deployment time, power use or uptime.
How far the first claim goes
The “first” in Dell’s headline carries a footnote. It rests on internal analysis by the Dell Office of the CTO, which compared fully integrated and validated, direct liquid-cooled networking racks from OEMs configured as part of an NVIDIA Vera Rubin NVL72 scalable production unit. Dell gives the reference date of that analysis as 11 November 2026. Readers should treat the ranking as Dell’s own comparison. The post does not say which other racks were included.
Dell also says PowerRack is a family of factory-integrated rack solutions spanning compute, networking and storage. This post covers only the networking rack.
What we do not know
Dell gives no price and no sales terms for the networking rack, and no availability for Europe. It does not say how many racks Nscale runs or when the deployment started, and the only customer it names is Nscale. Nscale does not comment in Dell’s post, and no one at the customer is quoted.
Sources
Written with AI from the makers' announcements and checked against the sources below. We did not test these devices.
Spotted a mistake? Let us know
This piece is based on the following public sources.