Why data centers are getting bigger: industry trend toward multi-gigawatt mega-campuses to optimize cost and power density.
Data centers were once considered large at hundreds of megawatts. The AI boom has transformed this entirely. Hyperscalers now build sprawling, city-sized campuses containing multiple data centers with vast computing power, alongside cooling infrastructure and on-site power generation—often necessary when the existing electrical grid cannot meet demand.
These multi-gigawatt facilities are proliferating worldwide to support AI workloads. A single 1 GW power plant, according to the U.S. Department of Energy, produces enough electricity for roughly 750,000 homes. To illustrate the scale: one planned facility in Utah, if fully operational, would consume more than twice the electricity the entire state currently requires.
The driving force is the nature of AI itself. Unlike smaller computing tasks such as serving webpages or hosting applications, training large language models demands massive, uninterrupted, low-latency computational clusters operating under one roof. Dell has noted that in LLM training, "the efficiency of distributed AI workloads is critical" and that network performance is essential—poor connectivity becomes a substantial bottleneck that "reduces hardware utilization, increases overall training time, and elevates operational costs." To train AI models efficiently, consolidating operations into a single colossal facility minimizes costs. Anthropic founder and CEO Dario Amodei stated in 2024 that training models "better than humans in most things" could cost $10 billion to $100 billion. In this competitive race, speed and capacity are paramount for companies like OpenAI, Google, Meta, and Anthropic.
Beyond training requirements, economies of scale drive expansion. Hyperscale data centers are far more efficient than traditional enterprise data centers. The International Energy Association reports that "the share of cooling systems in total data centre consumption varies from about 7% for efficient hyperscale data centers to over 30% for less-efficient enterprise data centers." Google, operating some of the world's most efficient facilities, confirmed that "hyperscale (meaning very large) data centers are far more energy efficient than smaller, local servers."
Mega campuses require abundant land and massive power access. High urban real estate costs push hyperscalers to rural areas where land is cheaper, and where they can build city-sized facilities with fewer nearby residents—often attracting interest through promised job creation.
The mega-campus era brings both advantages and liabilities. Economically, rural areas that lack existing data center infrastructure gain proven benefits. However, concerns persist: rising home energy costs, reduced quality of life near facilities, and pollution.
While these campuses plan to generate their own power and use efficient closed-loop cooling systems, the sheer scale of on-site power generation carries environmental consequences. As multi-gigawatt facilities burn natural gas to meet immediate demand, they release millions of tons of toxic gases. Stanford Law School quantified impacts for comparison: a 625 MW plant is projected to emit annually 90 tons of fine particulate matter (PM 2.5), 83 tons of nitrogen oxides (NOx), and 16 tons of sulfur oxides (SOx)—all documented contributors to respiratory and cardiovascular diseases, cancer, and adverse maternal and infant health outcomes. For multi-gigawatt facilities, multiply these figures by ten or more.
The scale is staggering: significant investments, substantial risks and rewards, impressive efficiencies offset by hefty environmental costs, and immense computing power. As AI development accelerates, infrastructure is being built that dwarfs traditional data centers and pushes computing to unprecedented limits. Regardless of one's perspective on this AI expansion, gigantic, city-sized megacampuses will soon emerge in rural towns worldwide.