NVIDIA's Vera Rubin is entering data centers, and the rise of agentic AI is accompanied by massive cost increases.

The Vera Rubin era has arrived. The first systems have been shipped to major cloud service providers, where they are undergoing validation and testing before large-scale deployment. With full production now underway, NVIDIA is anticipating even greater success than Blackwell.

However, as these systems are deployed, Foxconn Chairman Liu Yangwei noted that the cost of establishing gigawatt-scale data centers will grow exponentially. Vera Rubin will trigger significant transformation in the agentic AI space, delivering record-breaking computational performance that Blackwell cannot achieve.

According to Foxconn's data, AI data centers based on NVIDIA's Vera Rubin architecture could cost as much as $47 billion per gigawatt of installed capacity. Each data center will house up to 3,557 server racks, with each rack priced at approximately $9.1 million. The annual electricity costs for these AI data centers will approach $1.3 billion, with hardware depreciation costs reportedly six times the electricity costs.

Morgan Stanley Research's recent bill of materials analysis put the cost of VR200 NVL72 servers at approximately $8 million.

Meanwhile, construction costs are also extremely high. Liu Yangwei provided an example: building a 1-gigawatt AI data center centered on Vera Rubin requires capital expenditures of up to $47 billion and approximately 3,557 racks, with each Vera Rubin rack costing $9.1 million; additionally, the annual electricity costs for a 1-gigawatt AI data center total $1.3 billion, with hardware depreciation costs six times the electricity costs.

We are already seeing multi-gigawatt AI data centers under construction. By 2030, the global data center market is expected to reach $1.6 trillion, with global computing consuming 174 gigawatts of power—more than double the power capacity required in 2024 (68 gigawatts). This means that between 2025 and 2030, 18 gigawatts of new power capacity must be built annually to keep pace with demand growth.

The four major customers driving computational demand are AI model developers, cloud service providers, governments, and enterprises. Most of these customers are still in the early stages of AI enablement, but the future goal is to establish AI-native organizations, where all processes operate with AI at the core and humans simply set and manage objectives while overseeing workflows and results.

To this end, Foxconn's chairman proposed the idea of establishing "Taiwan-style" technology parks in the United States, primarily in Arizona and Texas. Related efforts are already underway, with these parks expected to take shape by year-end.

The Vera Rubin era marks a new frontier for AI, delivering unprecedented computational power for agentic systems, but at enormous cost. As the world races toward AI-native organizations and multi-gigawatt deployments, the coming years will test whether power capacity, capital, and innovation can scale in rapid coordination. The winners will be those who master not only the technology, but also the economics of building future AI infrastructure.