Texas has become the first major U.S. data-center hub to freeze new grid connections while officials determine how much of the extraordinary electricity demand attributed to the artificial intelligence boom actually exists.
The move follows an explosion in requests from data centers and other large electricity users seeking access to the Texas power grid. Those requests have risen from roughly 48 gigawatts in 2023 to more than 474 gigawatts today, according to a Sept. 1 Reuters report. That is more than five times the record peak electricity demand ever recorded on the Texas grid.
But the numbers are increasingly raising a basic question: How many of those proposed data centers are actually going to be built?
Across portions of the Midwest, Mid-Atlantic and South, very large power users—primarily data centers—have submitted requests totaling more than 700 gigawatts, Reuters found. That is more than 10 times the amount of electricity all U.S. data centers are estimated to consume today.
The enormous gap has prompted warnings about what regulators have begun calling “ghost demand”—projects that appear in utility connection queues but may be duplicated, speculative, underfinanced or otherwise unlikely to materialize.
For the best of our weekly content！
You are now signed up for our newsletter
Check your email to complete sign up
“When you don’t know what is real, you really don’t know how to build the infrastructure for it,” Texas Public Utility Commission Chairman Thomas Gleeson said at an industry conference, according to Reuters.
The AI boom is overwhelming traditional grid planning
The surge reflects the extraordinary infrastructure race surrounding artificial intelligence. Technology companies are spending hundreds of billions of dollars building the massive computing facilities required to train and operate increasingly powerful AI systems.
That race has also created an incentive to secure electricity before competitors do.
Developers, technology companies and landowners can seek grid capacity for proposed sites long before a finished data center begins consuming power. Utilities therefore face the difficult task of distinguishing serious projects from speculative proposals—and determining whether multiple requests may represent competing locations for what eventually becomes a single data center.
Texas illustrates the scale of the problem. In June, the Electric Reliability Council of Texas said it was already tracking more than 438,000 megawatts of proposed large-load connections, with nearly 89 percent coming from data centers. ERCOT’s historic peak demand, by comparison, is only 85,508 megawatts.
ERCOT has consequently abandoned its traditional practice of evaluating many large projects individually and introduced a “Batch Zero” system that examines major electricity users together so planners can determine how much demand the grid can realistically accommodate and what transmission infrastructure would be required.
“This new process represents a fundamental shift in how ERCOT manages the significant growth of large load interconnection,” ERCOT President and CEO Pablo Vegas said when the system was approved.
There are already signs that stricter scrutiny can make enormous amounts of projected electricity demand disappear.
Exelon reduced its estimate of high-probability data-center demand by approximately 40 percent, to 11 gigawatts, after imposing tougher collateral requirements, Reuters reported. In Ohio, AEP Ohio’s data-center pipeline fell by more than half after new rules imposed requirements including connection-study fees of as much as $100,000.
Those reductions suggest a significant portion of America’s apparent AI electricity boom may consist not of functioning server farms, but of projects whose developers are reserving a place in line.
The danger is building a grid for data centers that never arrive
The consequences extend well beyond technology companies.
Electric utilities must make investment decisions years in advance. If planners underestimate genuine data-center growth, there may not be enough generation and transmission capacity to meet demand. But if they assume every proposed project will be built, utilities could spend billions constructing infrastructure for customers that never materialize—and potentially pass those costs to households and other businesses.
The problem is already influencing electricity prices elsewhere. Data-center growth has contributed to sharply higher capacity costs in PJM Interconnection, the country’s largest regional grid, which stretches across 13 states and includes northern Virginia’s enormous concentration of data centers. PJM itself said in June that data centers are the “primary driver” of its electricity-demand growth and warned that new facilities can be constructed two to three times faster than much of the generation required to serve them.
Texas has now opted to force developers to show more of their cards.
Gov. Greg Abbott on Aug. 3 ordered a comprehensive audit of data centers seeking connections to the ERCOT grid. Projects must provide information allowing officials to verify whether they are legitimate before proceeding, and those failing to comply can be denied a grid connection.
The state is also demanding greater transparency over who ultimately owns proposed data centers, their water consumption, tax incentives and plans for on-site electricity generation. Similar concerns are spreading. Pennsylvania Gov. Josh Shapiro imposed tougher requirements in August for large data centers, while only 20 of more than 100 proposed projects in that state had actually sought the permits necessary to proceed, according to Reuters.
None of this means America’s AI-driven electricity boom is imaginary. Even after speculative projects are removed, utilities say confirmed data-center demand remains substantial enough to strain grids already struggling to add generation quickly enough.
Instead, the Texas crackdown highlights a different problem created by the AI investment frenzy: America’s power system is being asked to make multibillion-dollar decisions based partly on facilities that may exist only on paper.
Original article
Comments