텍사스 규제 당국은 지역 송전 용량을 압박하는 급증한 가상의 수요로 인해 신규 데이터센터의 전력망 연결을 일시 중단했다.

Texas has become the first major U.S. data center hub to freeze new grid connections while officials assess how much of the extraordinary electricity demand attributed to the artificial intelligence boom is real.

This decision follows a surge in interconnection requests from data centers and other large electricity consumers. According to a Sept. 1 Reuters report, those requests have climbed from roughly 48 gigawatts in 2023 to over 474 gigawatts today—more than five times the record peak demand ever recorded on the Texas grid.

Yet these figures raise a fundamental question: How many of these proposed facilities will actually be built?

Across parts of the Midwest, Mid-Atlantic, and South, large power users—predominantly data centers—have submitted interconnection requests exceeding 700 gigawatts, Reuters found. That volume exceeds ten times the estimated current electricity consumption of all U.S. data centers combined.

This massive discrepancy has prompted regulators to warn of “ghost demand”—projects listed in utility interconnection queues that may be duplicated, speculative, underfinanced, or otherwise unlikely to materialize.

“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 told an industry conference, according to Reuters.

This surge underscores the unprecedented infrastructure race driven by artificial intelligence. Technology firms are investing hundreds of billions of dollars to construct the massive computing facilities needed to train and operate increasingly powerful AI models. The competition has also created a strong incentive to secure grid capacity ahead of rivals.

Developers, tech companies, and landowners can apply for grid capacity years before a facility breaks ground. Consequently, utilities must distinguish viable projects from speculative applications and determine whether multiple submissions merely represent competing bids for a single eventual site.

Texas exemplifies the scale of the challenge. In June, the Electric Reliability Council of Texas (ERCOT) reported tracking over 438,000 megawatts of proposed large-load interconnections, nearly 89 percent originating from data centers. By comparison, ERCOT’s historical peak demand stands at just 85,508 megawatts.

To address this, ERCOT has moved away from evaluating large projects individually and implemented a “Batch Zero” framework. This approach evaluates major electricity consumers collectively, enabling planners to gauge realistic grid capacity and identify necessary transmission upgrades.

“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 upon approval of the system.

Early results indicate that stricter scrutiny can rapidly deflate projected demand figures. Reuters reported that Exelon cut its high-probability data center demand forecast by roughly 40 percent, down to 11 gigawatts, after implementing stricter collateral requirements. Similarly, AEP Ohio’s data center pipeline shrank by more than half following new regulations that included connection study fees reaching $100,000.

These reductions suggest that a substantial portion of America’s apparent AI-driven electricity surge consists not of operational server farms, but of speculative applications designed to reserve grid capacity.

The implications extend far beyond technology firms. Utilities must commit to multi-year investment cycles. Underestimating genuine data center growth risks insufficient generation and transmission capacity. Conversely, assuming every proposal will proceed could force utilities to spend billions on infrastructure for nonexistent customers, ultimately passing those costs onto residential and commercial ratepayers.

The issue is already impacting electricity pricing elsewhere. Data center expansion has driven sharply higher capacity costs across PJM Interconnection, the nation’s largest regional grid spanning 13 states and encompassing northern Virginia’s dense concentration of facilities. In June, PJM identified data centers as the “primary driver” of its electricity demand growth, warning that new facilities can be built two to three times faster than the generation capacity needed to support them.

Texas has now chosen to compel developers to substantiate their claims. On Aug. 3, Gov. Greg Abbott ordered a comprehensive audit of all entities seeking ERCOT grid interconnections. Applicants must submit documentation verifying project legitimacy; noncompliant projects risk denial of grid access.

The state is also mandating greater transparency regarding ultimate ownership, water usage, tax incentives, and on-site power generation plans. Comparable regulatory shifts are emerging elsewhere. In August, Pennsylvania Gov. Josh Shapiro enacted stricter requirements for large data centers, even as Reuters noted that fewer than 20 of the state’s 100+ proposed projects had secured the necessary permits to advance.

These measures do not imply that America’s AI-driven electricity surge is illusory. Even after filtering out speculative applications, utilities confirm that verified data center demand remains substantial enough to strain grids already struggling to expand generation capacity at pace.

Rather, Texas’s intervention highlights a distinct challenge born of the AI investment frenzy: America’s power infrastructure is being tasked with making multibillion-dollar capital decisions based, in part, on facilities that may exist only on paper.