Nvidia Pitches Water-Free AI Factories to Europe — and a Data Center That Arrives in 14 Days
Nvidia used its GTC Berlin keynote to announce that European ecosystem partners are adopting its DSX AI Factory Platform, which pairs closed-loop liquid cooling with dry coolers to nearly eliminate cooling water consumption. The company also unveiled DSX POD, a containerised Vera Rubin AI factory it says can cut deployment time from roughly six months to as little as 14 days, with global shipments starting in 2027.
Nvidia used its GTC Berlin keynote on 21 October to pitch European operators on two ideas that are really one: that an AI data center should be designed as a single system, and that it should arrive largely pre-assembled. The company said a group of European ecosystem partners is adopting its DSX AI Factory Platform, and it introduced DSX POD, a containerised reference design for complete Vera Rubin systems.
Designing the building around the chips
DSX is not hardware you can buy off a shelf. It is a reference architecture — plans plus software — that treats compute, networking, power delivery and cooling as one problem rather than four separate procurement decisions. Nvidia first showed the platform at GTC Taipei on 31 May 2026; the Berlin event was about turning it into something European cloud builders can actually order.
The water arithmetic
The headline claim concerns cooling. DSX reference designs pair closed-loop liquid cooling with dry coolers, so heat is rejected to air instead of being carried away by evaporating water. Nvidia says that at 2 gigawatts of deployment the approach can save up to 54 million litres of cooling water a day compared with conventional cooling towers — a figure it equates to the daily consumption of roughly 180,000 households.
There is a caveat built into the design. Dry coolers depend on ambient conditions, so on the hottest days a site may still need supplementary cooling, and local weather effectively decides how much. Nvidia also points to a software tool called MaxLPS, which it says can deliver the same AI output with as much as 30% less electricity.
Six months down to a fortnight
The more immediate commercial pitch is speed. DSX POD packages a full AI factory into modular containers that partners integrate, cable and validate at their own manufacturing facilities before shipping. Because the site only has to connect units that already work, Nvidia reckons the time from an empty building to serving tokens can fall from around six months to as little as 14 days.
Seventeen partners are named on the programme: Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies, Delta, Eaton, Foxconn (Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric, Vertiv and Wistron. Jensen Huang said the pods are intended for customers worldwide and will ship in 2027.
Europe's standing problem
The announcements land against an uncomfortable backdrop. Huang used the Berlin stage to acknowledge that Europe's AI industry lags the United States and China, and framed the new tools as a way to compress the buildout. That is a diplomatic sales pitch, but it also touches a real bottleneck: permitting, grid connections and available power — not silicon — are what usually delay a European data centre.
What the numbers do not say
Two things are worth watching. First, the water savings are stated as an upper bound at gigawatt scale; smaller sites, and those in hot climates, will see considerably less. Second, the 14-day figure measures installation, not the years of planning, land acquisition and grid approval that precede it. Nvidia has largely solved the assembly problem. Whether Europe can solve the power problem is a separate question, and it is one that no reference design answers.