The AI Data Center
What’s inside an AI data center, who builds and uses them, and what they mean for electricity, water and host communities.
AI data centers aren’t conventional data centers with more GPUs. Their architecture changes how compute, electricity, capital, water and communities interact.
This 17-page report follows that shift from the inside out: from rack-level design to who builds and finances these facilities, to their impact on the grid, water and host communities.
Key findings
Density is redefining the facility. Rack power is rising from 5–15 kW to about 135 kW, turning AI compute into an electrical-engineering problem.
Electricity is the binding constraint. Data centers could reach 9–17% of U.S. electricity by 2030, depending on which projects reach operation.
Enterprise AI isn’t building campuses. Much of it lands in colocation and cloud, which spreads grid and community impacts beyond the headline mega-sites.
Water accounting is incomplete. Much of the water footprint sits at power plants, not on site.
Written for investors, utilities and regulators, and host communities.