2025 is emerging as the defining year for deployment of gigawatt-scale AI data center projects globally.
The artificial intelligence industry is undertaking massive infrastructure expansions that defy conventional logic. xAI plans to expand its Colossus AI supercomputer from 100,000 GPUs to a million, which would cost $27 to $36 billion in hardware alone at typical pricing of $30,000 to $40,000 per GPU. This figure excludes the substantial costs for the building, cooling systems, and electrical infrastructure required to support the accelerators. The GPU nodes alone would demand approximately 1.2 to 1.5 gigawatts of electrical generation—more power than a typical nuclear reactor—and that accounts only for compute capacity.
Multiple tech giants are pursuing these gigascale AI projects. The Register identifies xAI, Meta, AWS, Oracle, and Microsoft as companies initiating or already advancing gigascale AI projects in 2025. Beyond the United States, India is racing to build its own gigascale AI infrastructure, with Reliance constructing a 1 GWh data center featuring Nvidia's Blackwell GPUs. Nvidia CEO Jensen Huang stated: "In the future, India is going to be the country that will export AI. You have the fundamental ingredients – AI, data and AI infrastructure, and you have a large population of users." China is also entering the competition, though specific details remain scarce.
The motivation driving this competition is the potential for dominance through artificial general intelligence. As one analyst notes, the nation or tech company that successfully creates AGI first "has a real shot at becoming the dominant power on Earth in perpetuity." The prize includes potential applications in drug discovery and reverse engineering human DNA, with possibilities for life extension or medical immortality for its developers. This perceived ultimate power explains the urgency with which governments and corporations are mobilizing resources.
Power generation remains the limiting factor, yet multiple solutions are emerging. While nuclear energy attracts substantial interest, fossil fuel-powered projects like Facebook's AI infrastructure demonstrate that nuclear delays will not constrain expansion. The article notes that eventually demand will exceed available nuclear and gas capacity, making new coal plants likely for powering AI data centers. China possesses a particular advantage: an economic slowdown has created a surplus of tens of gigawatts of electricity available for AI projects without new power plant construction. Combined with a large population of unemployed tech graduates and established second-tier chip fabrication capacity, China can substitute quantity for quality if US technology sanctions restrict cutting-edge chip access.
The speed of this infrastructure expansion has exceeded expectations. The author acknowledges surprise at the rapidity with which gigantic AI data centers—each consuming more power than a major city—have materialized, having anticipated at least a couple of years before the AI gold rush would reach this scale of activity.