Data centers already consume more electricity than many states. By 2030, AI infrastructure alone could demand as much power as the entire country uses today.
Sources: IEA Electricity 2024 report; Goldman Sachs AI Power Demand, 2024; EPRI Powering Intelligence, 2024; DOE.
TWh per year — bars show AI data centers specifically, not all data centers. US total annual electricity consumption is approximately 4,000 TWh.
% column = share of US total grid (~4,000 TWh). 2026–2030 are projections. Sources: IEA Electricity 2024; Goldman Sachs Global Investment Research.
Tracked facilities from our database. Capacity (MW) compared against each state's annual net generation. Water stress from WRI Aqueduct baseline (2023).
| State | Tracked facilities | Tracked capacity (MW) | State grid (TWh/yr) | DC load est. | Water stress |
|---|---|---|---|---|---|
| VA | 4 | 8,356 | 100 TWh | ~62.22% | Medium |
| IN | 5 | 5,640 | 117 TWh | ~35.89% | Low |
| WI | 4 | 5,320 | 68 TWh | ~58.25% | Low |
| LA | 1 | 5,000 | 94 TWh | ~39.61% | Low |
| TX | 7 | 4,500 | 495 TWh | ~6.77% | High |
| NE | 4 | 2,850 | 45 TWh | ~47.16% | Medium |
| OH | 5 | 2,250 | 150 TWh | ~11.17% | Low |
| PA | 3 | 1,920 | 246 TWh | ~5.81% | Low |
| MO | 2 | 1,900 | 90 TWh | ~15.72% | Medium |
| GA | 1 | 1,830 | 135 TWh | ~10.09% | Medium |
| MS | 3 | 1,800 | 44 TWh | ~30.46% | Low |
| NC | 2 | 1,270 | 127 TWh | ~7.45% | Medium |
| OR | 2 | 1,150 | 62 TWh | ~13.81% | Medium |
| AZ | 3 | 980 | 108 TWh | ~6.76% | Extremely High |
| NV | 2 | 780 | 41 TWh | ~14.17% | Extremely High |
| IA | 2 | 600 | 74 TWh | ~6.04% | Low |
| WA | 1 | 600 | 127 TWh | ~3.52% | Medium-High |
| UT | 1 | 400 | 44 TWh | ~6.77% | High |
| OK | 2 | 300 | 89 TWh | ~2.51% | Medium-High |
| IL | 1 | 200 | 208 TWh | ~0.72% | Low |
| SC | 2 | 200 | 99 TWh | ~1.50% | Medium |
| TN | 1 | 150 | 90 TWh | ~1.24% | Medium |
| NM | 1 | — | 27 TWh | — | High |
| MN | 1 | — | 62 TWh | — | Low |
| AL | 1 | — | 82 TWh | — | Medium |
| WV | 1 | — | 53 TWh | — | Low |
| MI | 1 | — | — TWh | — | Low |
| WY | 1 | — | 52 TWh | — | Medium-High |
Grid generation data from EIA State Electricity Profiles (2023). Water stress ratings from WRI Aqueduct Water Risk Atlas. Facility capacity figures are estimates from public filings and news. Load percentage = tracked MW × 8,760 hours ÷ state grid TWh × 100.
Utilities don't pass data center power costs directly to residents — but they don't absorb them either. When data centers sign massive long-term power purchase agreements, they lock up grid capacity and can accelerate infrastructure investment that gets recovered through rate cases.
In Virginia — which hosts more data centers than any other state — residential ratepayers have seen grid investment costs rise alongside the data center boom. Dominion Energy's rate filings have cited data center load growth as a driver of transmission and generation investment.
States with deregulated electricity markets (Texas, Illinois, Ohio) give data centers more power to negotiate rates that non-commercial customers can't access — effectively creating a two-tiered grid where industrial users pay less per kWh than households.
Communities near large data center clusters face a compounding problem: the facilities consume significant local water for cooling, generate noise from cooling systems, and can crowd out competing uses of industrial land — while employing relatively few local workers given their capital intensity. Residents rarely see economic benefits proportional to the resource burden placed on their grid, water systems, and infrastructure.
AI's power demand is now so large that Big Tech is buying nuclear power to run its data centers. The grid can't add clean, always-on power fast enough, so tech giants are restarting old reactors and funding new ones instead.
In 2024, Microsoft signed a 20-year deal with Constellation Energy to restart a reactor at Three Mile Island — the site of America's worst commercial nuclear accident. Amazon bought a data center next to the Susquehanna nuclear plant and invested in reactor startup X-energy. Google backed Kairos Power, and Meta put out a request for up to 4 gigawatts of new nuclear. Many of these reactors are small modular reactors (SMRs) that do not exist yet.
This matters for your bill and your community. Restarting and building reactors is expensive, and much of that cost can land on ratepayers and taxpayers through rate cases and subsidies. The reactors also raise old questions about waste, safety, and what it's like to live nearby.
We break the whole shift down in our guide to nuclear-powered data centers, including whether small modular reactors are safe, what they actually cost and who pays, and who is funding the nuclear-for-AI boom.
No federal law currently requires data centers to publicly disclose their power consumption. Voluntary ESG reports exist but are inconsistently defined and rarely facility-specific. Track pending US disclosure legislation in the AI legislation tracker.
Tell us what tools, data, checklists, or workflow would make your experience better.