NVIDIA has introduced DSX POD, a reference design that packs a complete Vera Rubin AI factory into modular containers built and tested by partners before they ship — cutting deployment at a prepared site from six months to as little as 14 days. Why it matters: the bottleneck in AI right now is not chips on a spec sheet, it is finished, powered, producing data centers — and NVIDIA is trying to turn building one into a manufacturing job instead of a construction project.
October 21, 2026 — Berlin
What happened
Announcing at GTC Berlin on October 21, NVIDIA said DSX POD units are complete AI factories assembled inside purpose-built containers by its partners. The computing, networking, cabling, rack-level power and liquid cooling are integrated and validated at the factory, so a unit arrives ready to connect to a prepared site’s power, cooling and networking out of the box.
That flips the usual build sequence. Instead of months of on-site assembly by specialized crews, partners finish the rack-to-rack cabling, wiring and piping and run functional validation before the container ever leaves the plant. NVIDIA’s pitch is that factory testing means fewer component delays and fewer costly on-site mistakes.
Who is building them
The partner list is long and heavy on established data center names: Dell Technologies, HPE and Lenovo on the systems side; Schneider Electric and Vertiv on power and cooling; plus Bull, Crusoe, Foxconn’s Ingrasys, GIGABYTE, Pegatron, Quanta Cloud Technology, Wistron, Eaton, Delta, DayOne, Armada and Anduril. NVIDIA says partners will deliver DSX POD AI factories worldwide in early 2027.
Founder and CEO Jensen Huang framed the launch around a simple metric: an AI factory only creates value once it starts producing tokens, not when construction wraps up. In NVIDIA’s vocabulary, “time to first token” — when a new facility starts running production workloads — is where deployment speed turns into revenue for operators.
Why it matters
Every hyperscaler and sovereign AI program is racing the same clock: demand for compute is here now, while conventional data center builds take quarters to years and compete for the same scarce electricians, cooling gear and grid connections. A containerized, pre-validated factory does not solve power supply — a site still needs power and cooling ready before a POD can light up — but it moves most of the integration risk off the critical path. If partners can genuinely deliver in early 2027 and hit anything close to the 14-day figure at prepared sites, expect the “AI factory in a box” model to become the default pitch for second-tier clouds, enterprises and governments that cannot wait out a traditional build. That is everything AI, tested against the calendar: the claim to watch is not the container, it is the stopwatch.
Frequently asked questions
What is NVIDIA DSX POD?
It is a reference design for complete NVIDIA Vera Rubin AI factories built by partners inside modular containers, integrated and validated before delivery to a customer site.
How fast can a DSX POD be deployed?
NVIDIA says factory-built units reduce deployment time at a prepared site from six months to as little as 14 days. The site still needs its power, cooling and networking ready first.
When will DSX POD be available?
NVIDIA says its partners will deliver DSX POD AI factories worldwide in early 2027.
Who are the DSX POD partners?
The named ecosystem includes Dell Technologies, HPE, Lenovo, Schneider Electric, Vertiv, Bull, Crusoe, GIGABYTE, Quanta Cloud Technology, Wistron and others.
Sources: NVIDIA via GlobeNewswire (October 21, 2026)