Nvidia wants AI factories built in real factories
The New Stack Frederic Lardinois ● Covered by 2 sources
Nvidia wants AI server blocks built in factories, then shipped to data centers ready to run. It could cut setup from months to 14 days, if the site has power, cooling, and network access.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Nvidia is pushing a more industrial version of AI infrastructure: build the rack system off-site, ship it in a container, and plug it in when it arrives. At its GTC Berlin conference on Wednesday, the company unveiled DSX POD, a reference design for shipping containers that partners fill with Vera Rubin racks, networking, and liquid cooling before they ever leave the factory.
The pitch is speed. Nvidia says conventional rack installation and wiring inside a data center can take months. A DSX POD, by contrast, can be ready to deliver tokens in as few as 14 days — but only if the destination already has the power, cooling, and network connections in place. Those are often the hard parts anyway.
Jensen Huang put the logic in plain terms: an AI factory makes money when it starts producing tokens, not when construction wraps up. Each container holds seven Vera Rubin NVL72 racks, or 504 Rubin GPUs. Ian Buck, Nvidia’s vice president of hyperscale and high-performance computing, said the unit packs about 10 million components and weighs around 28 metric tons. It also arrives with roughly 20 kilometers of copper cabling and 17 kilometers of scale-out networking already installed and tested.
Once it reaches the site, crews place the container, connect it to neighboring units, and run final burn-in tests. The cooling design is a big part of the story. Nvidia says Vera Rubin can run on coolant as warm as 45 degrees Celsius, which lets dry coolers do the job instead of the evaporative cooling towers common in many data centers today. In some climates, Nvidia says chillers may still be needed on the hottest days.
The company says that for a 2-gigawatt deployment, dry cooling could save about 54 million liters of water a day compared with cooling towers. And while containerized data centers are not new — Sun Microsystems, AWS, Microsoft, and others have all played with them — Nvidia’s version is less a mobile data center than a prefab building block for new AI factories. Armada is one of 17 partners building the systems, alongside Bull, Dell, HPE, Schneider Electric, and Vertiv, with worldwide deliveries due to start in early 2027.
My take — AI-written commentary, not fact-checked reporting
Nvidia has found the perfect way to sell plumbing as a breakthrough: make the boxes first, then call the boxes the future. That doesn’t make it fake — it just means the real bottleneck in AI is looking less like software wizardry and more like concrete, power, and pipes. Boring wins, which is deeply inconvenient for the hype business.
Read more about this at: The New Stack