ERCOT Data Centre Forecast of 228 GW by 2032 Tests Grid Rules Still Awaiting Final Form
Texas grid operator projects a more than thirtyfold rise in data centre load over six years against a reserve margin already narrowed by 1.9% demand growth.
Power availability, not processing capacity, has become the decisive constraint on AI data centre deployment across the United States, according to an industry analysis published on Wednesday (2026-07-16), with interconnection backlogs now pacing the buildout rather than capital or developer appetite. Texas is drawing concentrated demand from operators seeking grid headroom that other states cannot provide, placing ERCOT at the centre of a supply-adequacy question it has not faced at this magnitude before.5
ERCOT's load projection makes the scale concrete. The grid operator estimates data centre demand will grow from 7.4 gigawatts in 2026 to more than 228 GW by 2032. For context, ERCOT's all-time peak demand across every user on the Texas grid — homes, factories, offices — was 85.5 GW, set in 2023. The data centre forecast alone would exceed that record by a factor of more than two.1
The grid operator moved to address interconnection in early June. ERCOT approved two rule packages on Tuesday (2026-06-02): one establishing batch-review criteria for large users seeking grid access, the other targeting the cascade risk that officials have identified as a specific vulnerability. Data centres can drop thousands of megawatts of load in milliseconds when internal systems trip, potentially triggering frequency excursions that spread to substations across the network. Both packages remain in draft.3
Virginia supplied the reference case for why ERCOT is concerned. In July 2024, approximately 1,500 megawatts of data centre load disconnected unexpectedly, enough to supply 375,000 Texas homes at peak demand. ERCOT officials have been explicit that rapid load shedding at the scale the grid's own forecasts imply would present a fundamentally different reliability problem.1
Reserve margins are already compressing before that load arrives. NERC's summer reliability assessment found ERCOT reserves tightened from 34% to 29%, driven by roughly 1.9% demand growth with no corresponding increase in available supply. The agency flagged elevated potential for insufficient operating reserves under above-normal summer conditions — a June finding that preceded the most recent acceleration in data centre announcements.2
Near-term ERCOT real-time signals lean bearish. Wind and solar additions have outpaced incremental demand through the summer window, keeping prompt prices under downward pressure. NYMEX Henry Hub front-month was quoted at $2.89/MMBtu as of Saturday (2026-07-19), down from the $3.264 level on Monday (2026-06-22), when convergence of summer heat demand and LNG export flows was providing market support. Soft gas prices remove cost-push transmission through the gas generation fleet that might otherwise tighten real-time pricing.4
The short-term softness and the longer-range demand surge describe different time horizons under different supply conditions. What closes the gap is the interconnection queue, and that depends on how ERCOT's June rule packages are finalised. If batch-review procedures extend connection timelines, or if the stability requirements prove commercially restrictive, part of the 228 GW projection will simply not arrive on schedule. That outcome would ease near-term adequacy pressure but would simultaneously confirm that the supply constraint the Wednesday (2026-07-16) industry analysis identified as the AI sector's core infrastructure problem is real and binding in Texas.3,5
The stability provisions still in draft — specifically how large loads must behave during frequency or voltage events — are the terms on which data centre developers will base their interconnection commitments. Their final form, expected later in 2026, will shape ERCOT reserve margins through the remainder of the decade more directly than any individual demand forecast.3