AI Is Devouring America's Power Grid — and the Grid Is Not Ready
By Sanna the Weaver • Wed Feb 04 2026 • Science
The artificial intelligence industry has a power problem, and it is not a small one. The hyperscale data centers being constructed to train and run AI models consume electricity at a rate that is growing faster than new generation capacity can be brought online. Morgan Stanley projects a net US power shortfall of 9 to 18 gigawatts by 2028 — equivalent to the entire output of 9 to 18 nuclear power plants — driven primarily by AI compute demand. This is not a distant risk. Grid operators in Virginia, Texas, and Georgia are already seeing requests to interconnect new data center loads that exceed available substation capacity in entire regions. Northern Virginia: The World's Data Center Capital Under Strain Loudoun County, Virginia — "Data Center Alley" — hosts approximately 35% of the world's data center capacity, drawing over 3,400 megawatts of electricity and growing. Dominion Energy, the regional utility, has proposed
0 billion in new transmission and generation investments over the next five years specifically to meet data center demand, costs that will partly be passed to residential ratepayers. The PJM Interconnection, which manages the electric grid for 65 million people across 13 states and DC, reported in early 2026 that its interconnection queue — applications from projects seeking to connect to the grid — has reached a record 300 gigawatts. Most of those applicants will wait years for approval. Tech Companies Build Their Own Power Unable to wait for the utility construction cycle, the largest AI companies are pursuing direct power arrangements. Microsoft reactivated the Three Mile Island nuclear plant under a 20-year power purchase agreement with Constellation Energy. Google has signed contracts with multiple small modular reactor (SMR) developers, including Kairos Power. Amazon has invested directly in SMR startup X-energy. These deals reflect a structural shift: the largest technology companies are effectively becoming energy companies, with power infrastructure strategies as important as their compute strategies. "We used to think about the cloud as something invisible, something weightless. It is none of those things. It is concrete and steel and copper wire and generators burning fuel." — US Secretary of Energy, February 2026 The Climate Paradox The AI power surge is producing an uncomfortable paradox for the technology industry's climate commitments. Google, Microsoft, and Amazon have all pledged to be carbon-free or carbon-neutral by 2030 — goals that were calibrated against pre-AI-boom demand projections. By early 2026, all three companies' scope 2 emissions (from purchased electricity) are rising rather than falling, driven by data center expansion. Renewable energy procurement is accelerating, but the build cycle for large-scale solar and wind projects is 3 to 5 years; natural gas plants are being kept online and new ones planned to bridge the gap. The industry insists AI will ultimately help solve climate change. The near-term trajectory is pointed in the opposite direction.