NVIDIA Vera Rubin AI data centers could cost up to $47 billion per gigawatt as agentic AI demand surges
NVIDIA’s next major AI platform, Vera Rubin, is moving closer to large-scale data center deployment, and it could reshape the future of artificial intelligence infrastructure. The first systems are reportedly being shipped to major cloud service providers for validation and testing, with full-volume production already underway. After the huge momentum behind Blackwell, NVIDIA is now preparing for an even more ambitious era of AI computing.
Vera Rubin is expected to play a major role in the rise of agentic AI, a new class of artificial intelligence systems designed to complete complex tasks, make decisions, coordinate workflows, and operate with a higher level of autonomy. These capabilities require enormous computing power, and that demand is pushing the data center industry into a new phase where performance, power consumption, and cost are all scaling at unprecedented levels.
According to Foxconn Chairman Young Liu, building gigawatt-scale AI data centers based on NVIDIA’s Vera Rubin architecture could become extremely expensive. A single 1-gigawatt AI data center using Vera Rubin systems may require capital spending of up to $47 billion. Such a facility could include around 3,557 server racks, with each Vera Rubin rack estimated to cost roughly $9.1 million.
The cost of electricity is another major challenge. A 1GW AI data center could face an annual power bill of about $1.3 billion. However, electricity may not even be the largest ongoing expense. Hardware depreciation is estimated to be around six times higher than the yearly power cost, showing how quickly high-end AI infrastructure can lose value as newer, more powerful systems arrive.
Another estimate places the cost of NVIDIA VR200 NVL72 servers at around $8 million per unit, further highlighting how expensive next-generation AI hardware is becoming. These figures make it clear that the race to build AI data centers is not only about securing the fastest chips, but also about managing the massive financial burden of deploying and maintaining them.
The scale of future AI infrastructure demand is staggering. Multi-gigawatt AI data centers are already being planned and developed as cloud providers, AI companies, governments, and large enterprises compete for compute capacity. By 2030, the global data center market is projected to reach around $1.6 trillion. Global compute power demand could climb to 174GW, compared with roughly 68GW in 2024.
To support this growth between 2025 and 2030, the world may need to add about 18GW of new power capacity every year. That creates a major challenge for energy providers, governments, infrastructure developers, and technology companies. AI expansion is no longer limited by chips alone; it is increasingly limited by power availability, grid capacity, cooling systems, real estate, and construction speed.
The biggest customers driving this AI compute boom include AI model developers, cloud service providers, governments, and enterprises. Many organizations are still in the early stages of adopting AI, but long-term plans are becoming more aggressive. The goal for many companies is to become AI-native, where artificial intelligence is built into every workflow, operation, and decision-making process.
In that model, humans would focus more on defining goals, supervising outcomes, managing exceptions, and guiding strategy, while AI systems handle much of the execution. This shift could transform industries such as finance, manufacturing, healthcare, logistics, defense, research, and software development. However, it also means the demand for AI compute infrastructure will continue to rise sharply.
To support this expansion, Foxconn’s chairman has suggested building Taiwan-style science and technology parks in the United States, particularly in states such as Arizona and Texas. These hubs could help accelerate AI data center construction, semiconductor manufacturing, server assembly, and advanced technology development. Efforts to move these plans forward are expected to progress in the near future.
NVIDIA’s Vera Rubin platform represents more than just another generation of AI hardware. It signals a turning point for the entire data center industry. The future of agentic AI will require not only faster GPUs and more advanced servers, but also massive investment in energy, cooling, networking, construction, and long-term operational planning.
The companies that succeed in this new AI infrastructure race will not simply be those with access to the most powerful chips. They will be the ones capable of balancing performance with cost, power efficiency, supply chain control, and scalable deployment. As AI data centers grow from megawatt facilities into gigawatt-scale campuses, the economics of artificial intelligence may become just as important as the technology itself.
