Nvidia Unveils Modular Blueprint to Slash AI Data Center Deployment to Two Weeks

Nvidia ($NVDA) CEO Jensen Huang unveiled the DSX POD reference design in Berlin, framing artificial intelligence as a manufacturing industry and introducing modular containerized architecture designed to reduce AI factory deployment times from six months to as little as 14 days at prepared sites.
Each prefabricated container houses seven Vera Rubin NVL72 racks featuring 504 GPUs, alongside 20 kilometers of copper cabling and 17 kilometers of pre-installed networking. The architecture also incorporates warm-liquid cooling capable of operating at up to 45 degrees Celsius, enabling dry coolers to replace traditional evaporative towers and saving an estimated 54 million liters of water daily on a 2-gigawatt installation.
Huang highlighted that a 1-gigawatt AI factory requires roughly $50 billion in capital expenditure and can generate up to $40 billion in annual revenue, making rapid operational turnaround critical. While deployment speed remains constrained by external power-grid availability, 17 global manufacturing and infrastructure partners—including Dell Technologies ($DELL), Hewlett Packard Enterprise ($HPE), Foxconn, Schneider Electric, Crusoe, and Vertiv ($VRT)—have signed on, with worldwide deliveries targeting early 2027.