By Saf Malik
Senior Content and Insights Manager
Executive Overview
The global explosion of artificial intelligence (AI), cloud computing, and high-performance workloads has sparked a historic gold rush for electrical power. Across the United States, utility companies and grid operators have been inundated with staggering requests to hook up massive data centers to the local power supply. Headline figures routinely tout numbers exceeding 1,000 gigawatts (GW)—a staggering volume of electricity that rivals the total generating capacity of the entire United States power grid.
However, behind these eye-watering statistics lies a sobering reality. According to comprehensive new analysis from energy research firm Wood Mackenzie, roughly 768 gigawatts—or a staggering 72% to 76% of all current US data center interconnection requests—constitutes what experts term "phantom load."
In simple terms, three out of every four megawatts requested in official grid queues are speculative, duplicative, or dead-on-arrival proposals. Developers are shopping for the best regional incentives, land deals, and power-purchase agreements by filing the exact same project blueprints across multiple competing utility jurisdictions. While power companies and regional transmission organizations (RTOs) grapple with massive infrastructure backlogs, the true, realistic commitment pipeline sits at a modest 28% (comprising roughly 12% with firm service commitments and 17% deemed likely to proceed).
This profound divergence between perception and reality is distorting regional energy planning, inflating capital expenditure expectations, and triggering regulatory crackdowns from Texas to London. As grid operators struggle to distinguish legitimate multi-billion-dollar computing campuses from speculative filings, the energy transition faces an unprecedented administrative bottleneck.
Detailed Chronology: The Anatomy of an Interconnection Crisis
To understand how US electrical grids became choked with speculative data center demand, it is necessary to trace the convergence of the generative AI boom and legacy utility queue frameworks.
Phase 1: The Generative AI Gold Rush (2023–2024)
Following the widespread commercialization of large language models and generative AI tools in late 2022 and 2023, hyperscalers (such as Microsoft, Google, Amazon, and Meta) along with specialized AI neo-cloud providers began hunting for real estate capable of housing dense clusters of power-hungry graphics processing units (GPUs). Standard data centers historically consumed 10 to 30 megawatts (MW); the new generation of AI training facilities routinely demands 100 MW to upwards of 1 gigawatt per site.
Unable to wait for standard grid planning cycles, tech developers rushed to submit interconnection queue applications anywhere land was cheap and cooling was viable—disproportionately targeting the US South, Southwest, and major transmission hubs.
Phase 2: The Rise of Multi-Filing and "Phantom Load" (2024–2025)
As queue wait times stretched from months to several years, developers adopted aggressive hedging strategies. According to Ben Hertz-Shargel, Wood Mackenzie’s global head of grid edge, the pipeline became heavily skewed toward a small number of massive, speculative projects.
Real estate and infrastructure developers frequently submitted identical project proposals to multiple utilities and transmission operators simultaneously. By blanketing neighboring balancing authorities with duplicate applications, developers could shop for the most favorable connection timelines, local tax abatements, and energy pricing structures. While rational from a commercial risk-management perspective, this practice created a massive administrative illusion. Headline queue figures skyrocketed, giving the false impression that regional power grids would soon be completely overwhelmed by artificial intelligence workloads.
Phase 3: Regulatory Pushback and State-Level Audits (Mid-2025–2026)
The breaking point arrived when regional grids realized the physical impossibility of fulfilling these queues. In August 2025, the compounding strain of speculative demand in Texas prompted Governor Greg Abbott to order a comprehensive state audit of roughly 300 data center projects tracked by the Electric Reliability Council of Texas (ERCOT).
By mid-2026, the crisis had forced major utilities to actively prune their pipelines. Exelon, for example, aggressively trimmed its internal data center interconnection pipeline by nearly 40%, shedding speculative dead weight to settle on an 11 GW queue it considers operationally credible. Simultaneously, grid operators nationwide began overhauling their application rules, instituting stringent application fees, mandatory financial collateral, and credit-rating thresholds designed to price out unserious bidders.
Supporting Context & Metrics: Breaking Down the Numbers
The Wood Mackenzie data reported by Bloomberg provides a stark mathematical breakdown of the current state of US power integration.
- Total Requested Capacity: ~1,066 Gigawatts (GW) across the United States.
- Firm Service Commitments: ~12% (Projects that have signed binding agreements, secured land, and backed commitments with hard capital).
- Probable Projects: ~17% (Proposals with a high probability of reaching final investment decision and construction).
- Phantom Load / Speculative Queue: ~768 GW (~72% to 76%), consisting of duplicate filings, early-stage concepts, and land-banking maneuvers unlikely ever to break ground.
The Texas Microcosm: ERCOT’s 474 GW Dilemma
No single region illustrates the distortion of phantom load quite like Texas. ERCOT—the independent system operator managing most of the Lone Star State’s electrical grid—has tracked an astounding 474 gigawatts of data center interconnection requests.
To put this figure in perspective, 474 GW is more than five times the absolute all-time peak demand ever recorded on the ERCOT grid. If every requested data center were built, Texas would require power generation equivalent to nearly half the entire generating capacity of the United States. This astronomical mismatch between physical reality and queue data was the primary catalyst behind the state’s aggressive regulatory audit launched under Governor Abbott.
A Global Phenomenon: The UK’s Parallel Crisis
The disconnect between data center ambition and grid reality is not uniquely American. Similar patterns have materialized across Europe, where lax entry barriers to connection queues allowed developers to hoard capacity.
In the United Kingdom, investigative reports by The Guardian highlighted widespread "phantom investments," raising serious questions about the credibility of billions in pledged infrastructure. The UK’s energy regulator, Ofgem, was forced to step in with an aggressive power crackdown after national electricity connection applications ballooned from 41 GW to 125 GW in under a year, driven almost entirely by speculative data center and crypto-mining proposals. Ofgem subsequently introduced strict "use-it-or-lose-it" milestones to flush out speculative actors hogging transmission capacity.
Official Statements and Industry Insights
Industry leaders and analysts emphasize that while the AI data center boom is undeniably real, the administrative framework governing power distribution is dangerously outdated.
"The pipeline is weighted toward a small number of massive, speculative projects concentrated in the South and Southwest. Developers are submitting the same project to multiple utilities simultaneously while they shop for the best terms and site. The practice inflates headline demand figures without adding genuine buildable capacity."
— Ben Hertz-Shargel, Global Head of Grid Edge, Wood Mackenzie
Grid operators and federal regulators are increasingly alarmed that phantom loads are delaying genuine industrial electrification projects, renewable energy integration, and manufacturing expansions. The Federal Energy Regulatory Commission (FERC) has made large-load interconnection reform a top priority, pushing RTOs and ISOs to overhaul archaic first-come, first-served queue rules that reward speculation.
However, systemic reform is moving slowly. Notably, several regional grid operators opted to delay implementing FERC’s mandated large-load reform guidelines, signaling internal friction between traditional utility planning models and the breakneck speed of the tech sector.
Future Outlook: Navigating the Fog Until 2028
Looking ahead, energy analysts warn that the visibility problem plaguing utility queues will not resolve overnight. Wood Mackenzie projects that it will take until at least 2028 for the market to fully separate genuine, financed commitments from speculative noise.
Until then, headline interconnection metrics will remain a deeply unreliable barometer for future electricity demand. For utility executives, this uncertainty creates immense capital allocation challenges: build too much transmission infrastructure for phantom loads, and rate-paying consumers will be left footing the bill for stranded assets; build too little, and the region risks missing out on the economic engine of the next technological revolution.
For data center developers, the era of frictionless queue-stuffing is rapidly coming to a close. As grid operators enforce stricter financial guardrails—including non-refundable deposits, rigorous credit checks, and strict development milestones—speculative players will find it increasingly expensive to park fake projects in public power queues.
Ultimately, while the AI revolution will undoubtedly require vast quantities of new power, the sorting process currently underway will ensure that tomorrow’s grid is built for physical reality, not paper promises.
