Jensen Huang said at the GTC Berlin event on October 21, 2026, that a three-month delay in bringing a large AI data center online would cost roughly $10 billion in lost revenue. He estimates that a 1-gigawatt facility requires about $50 billion in construction spending and could generate around $40 billion in annual revenue once operational. He argues that the main bottlenecks are now grid connections and construction timelines, not just chip supply. Nvidia plans the global rollout of DSX POD, a modular, factory-built design integrating Vera Rubin systems, networking, and liquid cooling, with deliveries beginning in early 2027 through 17 partners. The advertised 14-day deployment applies only once the unit is connected to existing infrastructure, while Nvidia’s supply and capacity commitments reached $279 billion as of July 26, 2026.
Source: Read the original article
