Nvidia's DSX POD puts an AI data center inside a 28-ton shipping container, cutting setup from six months to 14 days once it arrives. But that's only at a "prepared site," and getting grid power in the US now takes about four years on average. You'll see how the box works, who it's really for, and what it means for AI's next bottleneck.

At its GTC event in Berlin on October 21, Nvidia unveiled the DSX POD, a design for an entire AI data center packed into a box the size of a shipping container. It arrives fully built and tested, and Nvidia says it can start running AI in as little as 14 days, instead of the usual six months. That sounds like the end of AI’s construction bottleneck. It isn’t, quite. The 14-day clock only starts once the site already has power, cooling, and fiber, and power is the hardest part to get. Here’s how the box works, why Nvidia built it, and where it will actually make a difference.

Think of a DSX POD as a data center that’s built in a factory and delivered on a truck, rather than assembled piece by piece on-site.

A single unit has roughly 10 million components and weighs about 28 metric tons (around 62,000 pounds). Because it uses a standard container size, it can travel by truck rather than needing special heavy transport.

Nvidia calls these “AI factories” because they produce tokens, the small chunks of text, code or images that AI models generate. When ChatGPT writes you a paragraph, it’s producing tokens. More GPUs mean more tokens per second, which means faster answers and more users served.

Nvidia announced the DSX POD on X with one line that sums up the pitch: deployment at a prepared site drops “from six months down to 14 days.”

AI factory deployment at a prepared site: from six months down to 14 days.

NVIDIA DSX POD brings Vera Rubin AI factories in containers, integrated and validated before delivery.

Announced at GTC Berlin: https://t.co/8MmpuCLSt5 https://t.co/HGsnP5FXos

— NVIDIA Newsroom (@nvidianewsroom) October 21, 2026