NVIDIA has introduced DSX POD, a modular AI factory design that can be installed at a prepared site in as little as 14 days. AI infrastructure is beginning to look less like a construction project and more like a manufactured product.

[For more news, click here]

Businesses are racing for AI infrastructure; the models and chips matter, but they add little value if it takes months to build the data center needed to run them. NVIDIA has introduced DSX POD to compress that timeline. It is a modular reference design for complete Vera Rubin AI factories that are assembled, integrated and validated by partners before being delivered to customers.

According to NVIDIA, a DSX POD can be brought online at a prepared site in as little as 14 days, compared with around six months for a more traditional deployment. NVIDIA Newsroom

This number matters because it targets one of the biggest bottlenecks. Building the compute is important, but getting it operational is another matter. Traditional AI infrastructure requires complex installations. A DSX Pod moves most of the installation work off-site.

Partners assemble and validate the systems inside modular containers before delivery, including rack-to-rack cabling, wiring, piping, networking and cooling integration. Customers still need to have a site with required power, cooling and network connections. But off-site integration can run in parallel with site preparations, significantly shortening the timeline.

As NVIDIA CEO Jensen Huang put it, an AI factory begins creating value when it produces tokens, not when construction finishes. By moving more integration work into controlled manufacturing facilities, NVIDIA argues that operators can reduce on-site errors, component delays and specialized labour requirements. For AI cloud providers, governments and enterprises, that could make capacity expansion more predictable.

This can also apply to other AI factories that are traditionally created like bespoke construction programs; operators can deploy repeatable modules as per demand.

NVIDIA is not doing this alone, they are working with a large group of computing and infrastructure partners including Dell Technologies, HPE, Lenovo, Bull, Schneider Electric, Vertiv, Eaton, Foxconn, GIGABYTE, Pegatron, Quanta Cloud Technology and Wistron. Systems are expected to become available globally from early 2027. NVIDIA Newsroom

At the same GTC Berlin event, the company also announced that European partners are also adopting DSX designs aimed at reducing water and energy requirements.

NVIDIA says closed-loop liquid cooling and dry-cooling systems can reduce facility cooling-water consumption dramatically, while its DSX MaxLPS software can help operators generate the same AI output with up to 30% less power. These numbers are important as energy availability and grid constraints are a challenge.

This development could industrialize AI factories and they can be manufactured, tested and deployed in weeks rather than months. If it is successful it is a scalable industrial product.

Related Articles

SUSE and NVIDIA Target the Governance Gap in Enterprise Agentic AI

Salesforce Buys Braze as Real-Time AI Reshapes Customer Engagement

Microsoft’s Maia 300 Chip Targets the Cost of Running AI