Jensen Huang Puts a $50 Billion Price Tag on a 1GW AI Factory at Nvidia’s GTC Berlin

Nvidia used its GTC Berlin keynote to sharpen the focus on the cost and scale of building AI capacity, with CEO Jensen Huang putting a $50 billion price tag on what he described as a 1 gigawatt, or 1GW, AI factory. Speaking at the event, Huang estimated that such a facility would include not only computing equipment, but also the land, power supply, networking, storage, cooling and other infrastructure required to run it.

The remark reframes AI infrastructure from a software-and-chip story into a full systems-and-site project, where power availability and physical buildout are as consequential as the accelerators themselves. A 1GW scale facility implies a massive electrical and cooling footprint, meaning timelines and permitting constraints can matter as much as procurement schedules for servers and networking gear.

Huang also tied the infrastructure discussion to Nvidia’s push for faster deployment. In the same keynote context, he referenced the launch of DSX POD, described in the reporting as a new “pod” designed to help deliver AI compute and related infrastructure in packaged form rather than as a fully custom build from scratch.

The DSX POD framing suggests Nvidia is trying to reduce the friction between demand for AI capacity and the realities of building data-center environments capable of supporting high-density computing. In practical terms, packaging can affect how quickly a customer moves from planning to installation, how predictably systems are configured, and how repeatable deployments become across multiple sites.

For investors and customers alike, Nvidia’s choice to talk about total facility costs is notable because much market attention tends to center on unit economics of chips, software, and networking. A $50 billion number highlights that the biggest bottlenecks may be external to the chip itself, such as grid power connection, cooling engineering, and on-site construction.

Still, the company has not disclosed additional specifications or commercial terms for DSX POD in the reporting that reached the market. There is no public detail in the available material on what configurations are included, how pricing is structured, or how the pod integrates with specific customer power and facility constraints beyond the general reference to its role as a new deployment building block.

Analysts will likely watch whether Nvidia provides more granular breakdowns in follow-on materials, such as guidance on deployment timelines, required power availability, and the typical components included in a pod-based factory layout. Those details would help quantify how much of the infrastructure cost is attributable to Nvidia’s systems versus customer-managed components.

What remains unclear from the current disclosure is whether Huang’s $50 billion figure represents a specific customer scenario, an industry-average planning assumption, or an illustrative benchmark for capital planning. Without a cost model or methodology, the number should be treated as a high-level estimate rather than a precise budgeting template for every buildout.