AI Power Demand Is Breaking the Grid Faster Than Anyone Is Building It. Here Is Where Things Stand.

Since Unbiased Headlines covered the grid performance problem on June 23, three developments have added hard numbers to the diagnosis: a 2.67 GW dedicated gas plant deal between Chevron and Microsoft in West Texas, updated Lawrence Berkeley National Laboratory electricity consumption projections, and a published analysis from Columbia researchers quantifying the cost pressure landing on ordinary ratepayers.

The Chevron-Microsoft Deal

Chevron announced that its wholly owned subsidiary Energy Forge One has signed a 20-year power purchase agreement with Microsoft to build a co-located gas generation facility in West Texas, according to Power Engineering. The project, called Project Kilby, targets approximately 2.67 GW of capacity built in phases, primarily through large GE Vernova turbines with additional capacity from Caterpillar subsidiary Solar Turbines. First power delivery is anticipated in 2028.

The design bypasses the existing transmission grid entirely, delivering electricity directly to Microsoft's data center. Chevron says surplus power could eventually be sold back to the grid through future interconnects, but that is a secondary objective, not the primary design goal. Non-potable brackish groundwater replaces freshwater for plant operations, and selective catalytic reduction systems are included to reduce NOx emissions.

"AI is reshaping the global economy, and abundant, affordable, reliable energy is essential to fueling that transformation," said Jeff Gustavson, Chevron's president of New Energies. The project is the first concrete output of a partnership Chevron, GE Vernova, and Engine No. 1 announced in 2025 targeting up to 4 GW of dedicated data center power.

What It Costs Ratepayers

Data centers consumed 4.7% of all U.S. electricity in 2024. Lawrence Berkeley National Laboratory researchers now project that share climbing to between 9.5% and 15.3% by 2030, according to Utility Dive. That is a meaningful upward revision from LBNL's own forecast two years ago, which put the 2028 figure at 6.7% to 12%.

The cost signal is already visible. Columbia University researchers cited in the same Utility Dive report noted that average residential electricity prices tracked inflation between 2019 and 2024, then jumped 6% in nominal terms last year—more than twice the inflation rate. Investor-owned utilities filed $18 billion in rate increase requests in 2025, the most since the mid-1980s, according to the LBNL data the Columbia team referenced. Regulators approved 66% of the dollar value requested.

The Columbia researchers identified the incentive structure as part of the problem. Under traditional utility regulation, utilities earn roughly a 9% to 10% return on capital investment. "This regulatory structure rewards capital deployment more than system optimization," they wrote. The fix they propose—grid-enhancing technologies and demand response—is commercially available now. Replacing existing transmission lines with advanced conductors alone could generate $180 billion in savings by 2050, per an analysis in the Proceedings of the National Academy of Sciences.

The Infrastructure-vs.-Generation Distinction

Amanda Simonian, chief marketing officer at TerraFlow Energy, published a piece in Utility Dive arguing that the sector is misdiagnosing the problem. More generation does not automatically fix congestion at constrained transmission nodes, instability from volatile load behavior, or local system stress when large loads cluster faster than infrastructure can adapt. PJM Interconnection has already flagged reserve margin concerns; ERCOT is tracking rising demand scenarios; the Electric Power Research Institute projects major increases in data center consumption.

The strongest counter-argument deserves fair treatment: the co-located model Chevron and Microsoft chose is a direct response to exactly this critique. By building generation physically at the load site and bypassing the transmission grid, Project Kilby sidesteps the congestion problem rather than solving it. But it also avoids making it worse for existing ratepayers. Critics of that approach will note it fragments grid planning and makes regional reliability harder to manage over time, since isolated industrial loads eventually do interact with the broader system as they scale.

China's Nuclear Advantage Is Widening

While the U.S. builds gas plants to feed AI data centers, China is building nuclear. OilPrice.com reported that since Georgia's Plant Vogtle came online in 2024, zero new U.S. nuclear reactors have broken ground. In the same window, China added 34 GW of nuclear capacity and now accounts for nearly half of all reactors currently under construction worldwide.

"By a wide margin, China will have the world's most dynamic and significant nuclear industry through 2035," Damien Ma, energy lead analyst at Gavekal Technologies, wrote in a recent report cited by the South China Morning Post. China builds a new plant in roughly six years; Vogtle took more than a decade and ran billions over budget.

President Trump has stated an ambition to achieve "lasting American dominance in the global nuclear energy market" and his administration has moved to reduce regulatory barriers. Whether deregulation can compress U.S. construction timelines enough to change the trajectory is the unresolved question. OilPrice.com notes the obvious tension: nuclear is a high-consequence industry, and cutting oversight to accelerate timelines carries real public safety risk that any honest accounting has to include.

The Open Question

The Chevron-Microsoft deal and the co-located generation model provide a near-term answer to data center power demand. Whether that model scales without eventually creating new grid stability problems, and whether the $180 billion in potential transmission savings from advanced conductors gets captured before ratepayer bills climb further, are the two questions that will define the next phase of this story. The Columbia researchers have called for structural regulatory reform to change utility incentives. No state or federal regulator has committed to that path as of June 23, 2026.