The Rational Return of Data Center Site Selection: The Cautious Expansion Logic Advocated by AI Itself

When AI Becomes Its Own Site Selection Consultant

In the AI-driven data center construction boom, an unexpected voice has joined the discussion: AI itself. Newsweek posed the same question to mainstream AI systems such as Microsoft Copilot, Google Gemini, ChatGPT, and Claude—"Where should data centers be built?" The answers were not only consistent but also cautious: rather than "anywhere, as fast as possible," priority should be given to areas with sufficient electricity and water resources, avoiding environmentally sensitive zones, and conducting adequate community consultation.

This result is quite ironic: the very technology driving the surge in demand for data centers advocates a more restrained path of expansion. This provides an unexpected reference framework for infrastructure investors and engineering planners—when community opposition to project locations is growing, AI's "rational site selection" could become a tool to mitigate conflict.

Geographic Logic: From Power Corridors to Industrial Brownfields

All the AI models queried pointed to similar site preferences: existing power generation corridors, areas already zoned for industrial use, and retired coal-fired power plants. Microsoft Copilot explicitly stated: "Data centers should be built in places that impose the least burden on existing communities, not where land is cheapest or tax incentives are greatest." It suggested converting abandoned steel mills into data centers rather than occupying farmland or residential areas.

Google Gemini further refined this logic: high-density AI training clusters should be located in states such as Minnesota, Wisconsin, New York, and Michigan; while large-scale cloud infrastructure should be integrated into retired coal-fired power plants in Pennsylvania, Ohio, West Virginia, Nevada, Utah, and Idaho. This site selection strategy is essentially infrastructure reuse—leveraging existing transmission networks and water permits to reduce environmental assessment time for new land acquisition while avoiding direct conflict with communities.

For engineering capital, this means the geographic center of gravity for data center investment may shift away from traditional areas like Northern Virginia and Texas. The Midwest and Great Lakes regions, with their cooler climates (reducing cooling energy consumption), relatively abundant water resources, and existing industrial land, are emerging as new growth poles.

Resource Constraints: Electricity and Water Become Core Bottlenecks

AI models generally emphasized the inescapability of two key resources: electricity and water. ChatGPT suggested that "data centers should be built where energy is most abundant, not where energy is cheapest," and specifically warned against areas with severe water shortages, dense suburbs, high-quality farmland, flood-prone coastal zones, and heavily constrained power grids. Claude stated bluntly that most of the southwestern United States should not be considered for new data centers due to water stress.This warning aligns with real-world data. According to a crowdsourced map initiated by Erin Brockovich, community complaints about data centers across the U.S. focus on electricity consumption, water usage, noise, and approval speed. In places like Virginia and Texas, the electricity demand of data center projects has strained local grids and driven up residential electricity rates. For project financiers, long-term power purchase agreements (PPAs) and water permits are becoming more critical due diligence items than land costs.

Community Gaming: New Rules for Transparency and Compensation

AI models repeatedly mentioned community engagement and transparency in their responses. Claude recommended conducting independent studies on water use and grid impact before construction, and publishing the results; implementing real penalties for noise and light pollution. It also stated bluntly: “Construction work is temporary, and permanent operational jobs at each hyperscale facility typically number only a few dozen to a few hundred. Local governments should negotiate incentives based on realistic long-term revenues, not the inflated forecasts companies often propose.”

This reflects the rising risk of social license for data center projects. In Caledonia, Wisconsin, Microsoft abandoned a data center plan due to strong community opposition. Even though Microsoft committed to paying rates sufficient to cover electricity costs, minimizing water use, and “replenishing more water than consumed,” it could not fully dispel local concerns. In the future, data center developers may need to accept stricter environmental impact assessments, community benefit agreements, and performance-linked tax breaks.

Long-Term Trends: The Governance of AI Infrastructure

When AI itself suggests that data center siting should be more rational, more dispersed, and more focused on brownfield redevelopment, this goes beyond a mere technical discussion. It signals that AI infrastructure is moving from pure engineering expansion into the governance phase. National governments, local governments, utility companies, and communities will jointly define the rules for siting data centers.

• For global infrastructure observers, this trend is illuminating:
• In developed countries, data centers will increasingly be tied to industrial brownfield redevelopment and regional economic transformation (e.g., former steel mill sites in Pittsburgh).
• In developing countries, data center siting may face similar water-power conflicts, but policy incentives and labor costs will still drive growth.
• In terms of capital structure, ESG requirements and long-term operational risks will push more data center projects to adopt PPP models and incorporate community participation into financing conditions.

Newsweek's experiment provides a unique “system feedback”: the siting logic of the infrastructure driving AI computing should not be led solely by cost and speed, but should incorporate long-term considerations of resource carrying capacity and community well-being. For engineers, investors, and policymakers, this may be the most practical advice AI has to offer.*Note: This analysis is based on the report "We Asked AI Where Data Centers Should Go—The Answers May Surprise You" published by Newsweek on July 21, 2026.*