Texas
ERCOT Warns Texas AI Power Boom May Not Materialize
Texas is planning its grid around an unprecedented wave of AI-driven power demand that the state’s energy regulator says may not fully materialize on projected timelines.
In a recent filing to the Public Utility Commission of Texas, the Electric Reliability Council of Texas (ERCOT) projected statewide power demand could surge to nearly 368 GW by 2032 – more than four times the state’s current peak demand record of 85.5 GW. But the filing also contains an unusual warning from the grid operator itself.
“ERCOT has concerns with using the preliminary load forecast values for the Reliability Assessment and any other transmission and resource adequacy analysis,” the organization wrote in its April 2026 long-term load forecast filing.
The organization added that it may seek adjustments to the forecast based on “actual historical realization rates or other objective, credible, independent information.”
ERCOT has already begun adjusting for realization risk internally. In its 2025 long-term load forecast report, the grid operator said the “average peak consumption per site was 49.8% of the requested MW” and applied that factor to projected non-crypto data center load additions in some planning models.
ERCOT President and CEO Pablo Vegas said the forecast reflects “higher-than-expected future load growth” tied to changing large-load planning dynamics.
Texas has emerged as a hotspot for data center growth, with numerous new projects reshaping the energy market and challenging grid capacity. (Image: Alamy)
Texas Developers Race Ahead of Grid Capacity
Texas has emerged as a key data center market, driven by its abundant land, competitive energy prices, and favorable regulatory environment. This combination has positioned the state as a magnet for hyperscale operators and AI infrastructure investments. The state is estimated to account for around 15% of all data center connectivity in the US.
Recent and proposed AI data center campuses tied to OpenAI, Oracle, Meta, Crusoe, CoreWeave, Soluna, and other hyperscale operators are reshaping Texas grid planning. Developers have proposed large campuses across North Texas, Abilene, West Texas, and the Houston corridor, many requiring hundreds of megawatts of capacity and, in some cases, dedicated onsite generation to bypass interconnection delays. That buildout pushed ERCOT’s non-crypto data center forecast above 228 GW by 2032.
Developers are continuing to pursue Texas aggressively because ERCOT still offers faster timelines and more flexible market structures than many competing regions. Several proposed campuses pair AI infrastructure with onsite gas generation, colocated power assets, or flexible-load arrangements to navigate mounting transmission constraints.
Utilities across the US are grappling with AI-driven electricity growth, but ERCOT’s projections stand apart for both scale and uncertainty. PJM Interconnection, the nation’s largest grid operator, expects summer peak demand to climb above 241 GW over the next 15 years as data centers and electrification expand. ERCOT, by contrast, projects demand potentially reaching nearly 368 GW by 2032, driven largely by proposed non-crypto data center loads. At the same time, the grid operator openly questions how much of that demand will materialize on schedule.
Bigger Than Texas
Similar pressures are emerging elsewhere. In California, CAISO’s latest transmission plan cited “data center load growth” as a driver of major grid upgrades and described interconnection volumes as “unmanageable” before recent queue reforms.
A recent Grid Strategies report reached a similar conclusion nationally, warning that the “data center portion of utility load forecasts is likely overstated by roughly 25 GW” compared with market-based deployment estimates.
Ihab Osman, an independent strategist specializing in data center and other mission-critical infrastructure, said the distinction is less about “real” versus “fake” AI demand and more about “announced versus deliverable demand.”
“A large share of the current AI/data center planned load should be treated as paper megawatts until it is validated through physical gates,” Osman said, citing factors including site control, transmission deliverability, generation availability, turbine and transformer supply, permitting, financing, and credible energization schedules.
Osman said ERCOT’s forecast is best understood as “a stress-test map, not as a fait accompli build map.”
Separating ’Paper Megawatts’ From Real Demand
The filing shows Texas regulators and grid planners struggling to distinguish operating AI infrastructure from a rapidly expanding pipeline of proposed projects.
“The vast majority” of ERCOT’s projected load growth comes from submissions provided by transmission and distribution utilities, according to the filing. Those requests include hyperscale AI campuses, GPU clusters, and other large industrial loads seeking future grid capacity reservations.
Alison Silverstein, a former senior adviser to the chairman of the Federal Energy Regulatory Commission, said “a large proportion” of projects in ERCOT’s large-load interconnection queue have already been canceled, particularly among smaller developers facing long interconnection delays and high turbine and transformer costs.
Forecasts Collide With Physical Infrastructure Limits
ERCOT has also signaled that many projects may not materialize on the timelines shaping transmission planning.
The grid operator said summer 2026 peak demand is likely to land between roughly 90.5 GW and 98 GW – far below the preliminary 112 GW figure embedded in the long-term forecast. ERCOT said it appears “unlikely” that new large-load projects and existing site expansions will ramp quickly enough to push demand that high this year.
The filing suggests uncertainty around AI-related load growth is beginning to influence broader infrastructure planning assumptions. By 2032, ERCOT projects non-crypto data centers reaching 228 GW of demand, compared with just 9 GW from cryptocurrency mining and roughly 3 GW each from hydrogen/e-fuels and oil-and-gas-related industrial growth.
The move also suggests the regulator is no longer simply forecasting AI-driven growth, but also working to determine how much of the proposed boom can actually be financed, supplied, interconnected, and energized before utilities commit billions to long-lived infrastructure.