An unprecedented wave of construction for artificial-intelligence data centers is straining the nation’s ageing electricity network and prompting technology companies to seek their own sources of power. Goldman Sachs estimates that some individual interconnection requests now reach 5 GW—enough to supply about five million homes—while Gartner projects total U.S. data-center demand could hit 500 TWh by 2027. Grid operators warn the system, much of it built in the 1970s, is approaching a “critical bottleneck.” The pressure is leading companies such as Google, Meta, Microsoft and Amazon to finance natural-gas plants, revive dormant nuclear projects and lobby the White House for faster permitting. A draft “AI Action Plan” circulated by the administration calls for streamlined approvals and exemptions from parts of the Clean Water and Clean Air Acts to accelerate construction of new generating capacity and related infrastructure. Regional flashpoints are already emerging. In northern Nevada—now one of the world’s largest data-center markets—developers have asked utility NV Energy for almost 6 GW of additional capacity, a level that would force the state’s power sector to expand roughly 40 percent. Water use is also a concern in the drought-prone area, where environmental groups and tribal leaders say new facilities could consume billions of gallons annually. With data-center footprints continuing to rise and energy costs climbing in high-growth regions, regulators face mounting questions over who will pay for grid upgrades and how to balance economic development with environmental commitments. Absent large-scale investment and regulatory relief, analysts say electricity shortages could become an impediment to the AI industry’s expansion in the United States.