An AI data center is a power-hungry building of GPU servers that trains and runs AI. 51 questions answered with sources: power, water, chips, costs, where they’re built, space, neighbors, rules.
Alphabet, Amazon and Meta plan about $545-570 billion of 2026 capital spending, double 2025. Who builds, the bubble debate, federal and state rules, global impact and a win-win plan.
Loudoun County gets 40% of its budget from data centers, but a typical Virginia site needs under 50 permanent workers. Taxes, jobs, water, noise and the backlash, with sources.
Musk says space will be the cheapest place for AI within 2-3 years. SpaceX’s May 2026 SEC filing says “as early as 2028” and it “may not achieve commercial viability.”
As of Oct. 2, 2026: Meta plans 5 GW in Louisiana, OpenAI’s Stargate has 7 US sites, Texas has 474 GW of grid requests. A state-by-state map with sources.
One AI rack draws about 120 kW. A 1 GW campus equals about 812,000 US homes. US data centers used 4.7% of the nation’s power in 2024, heading to 11.8% by 2030. GPUs, HBM and water, sourced.
An AI data center is built power-first: about 18 months to build 50 MW, 4+ years to get grid power in big markets, and $17.6M per MW before chips. The 7 steps, with sources.