Rahul Somvanshi
Electricity demand is climbing again in the United States, and data centres are a major reason. The change is most visible in the grids serving Texas and the Mid-Atlantic, where large computing facilities are seeking connections at a scale that can reshape local planning. For households and businesses, the question is not whether a single AI prompt causes a blackout. It is whether new demand is matched with enough generation, wires and transparent cost allocation.
The U.S. Energy Information Administration says national electricity demand grew by about 1.7% a year from 2020 to 2025, after averaging 0.1% a year from 2005 to 2019. Its February 2026 outlook projects annual load growth averaging 10% in ERCOT and 3% in PJM from 2025 to 2027. ERCOT covers most of Texas; PJM serves all or parts of 13 states and Washington, DC.
The immediate consequence is local. A new data centre needs a firm connection, substations and often transmission upgrades before it can operate at full scale. Those projects can take years. Regulators and grid operators must decide which costs belong to the customer requesting the connection and which are spread across other customers. That decision can affect bills even for people who never use the facility’s services.
Why data-centre load is testing the grid
Data centres are not the only source of rising demand. The EIA also points to expanded industrial electricity use. But the agency identifies data centres as a driver of the recent increase, particularly in ERCOT and PJM. This matters because grid operators must keep supply and demand balanced every second, including on the hottest afternoons and coldest mornings.
In an EIA scenario where demand grows faster than expected while future generating capacity stays unchanged, more electricity comes from existing fossil-fuel plants. The agency projects U.S. natural-gas generation would rise 7.3% from 2025 to 2027 in that high-demand case, compared with 1.7% in its February baseline. It also finds that coal supplies more than half of the additional generation in the PJM, MISO and SERC regions in the scenario because some coal plants have spare capacity. That is a modelling result, not proof that a particular AI facility has kept a named plant open.
The planning problem is broader than generation. Transmission lines, transformers and substations have to be built or upgraded to move electricity safely. The debate overlaps with the water and land pressures documented in Karmactive’s report on data-centre water demand in Spartanburg. The effects vary by site, cooling design, climate and the local power system.
Can data centres raise electricity prices for other customers? They can create costs that regulators must allocate, but the outcome is not automatic. A grid operator may require the project developer to fund dedicated equipment, while broader network upgrades can be handled differently under state and regional rules. The practical safeguard is public disclosure of projected load, required upgrades and who pays for them.
The key test is whether power-system investment arrives at the same pace as computing demand. Faster efficiency gains, flexible demand and new clean generation can reduce pressure. Without them, operators may rely more heavily on existing gas and coal capacity during tight periods. The EIA’s scenarios are forecasts, not a verdict on what regulators and utilities will choose.
