The CEO of PJM Interconnection, the largest grid system operator in the United States, has stated that extensive upgrades are necessary for both the PJM system and the wider national grid to handle unprecedented electricity demands driven by the construction boom of hyperscale AI data centers. Market research reports indicate that global power requirements for AI data centers are surging exponentially, far surpassing the flexible growth patterns of traditional IT loads. Over the coming years, U.S. utility companies and grid builders may need to invest trillions of dollars to modernize aging infrastructure and support the massive expansion cycles of tech giants like Microsoft, Google, and Amazon.
In an internal letter to stakeholders, David Mills highlighted that under the current grid structure—which serves over 67 million people across 13 states—PJM cannot ensure adequate power supply while shielding residential consumers from soaring bills. "The current situation is clearly unsustainable," Mills wrote in the letter released Wednesday. He added that visible strains on pricing, reserve margins, and investment pipelines reflect issues more fundamental than mere design recalibration.
Key pressures on PJM include anticipated severe power shortages as early as next year, necessitating immediate grid upgrades, and the potential exit of American Electric Power Co., one of the nation's largest utilities, from the system. Consequently, the electric utility stocks best positioned to benefit from this rally are those capable of delivering power quickly, reliably, and cost-effectively to data centers. These include regulated utilities with exposure to high-load growth areas, beneficiaries of transmission network upgrades, providers of rapidly deployable natural gas/nuclear/distributed power equipment, and data center power chain companies offering flexible load management and demand response capabilities.
Grid modernization is urgent. While the Trump administration has urged tech giants to adopt "self-powering" measures, such efforts are unlikely to alleviate rising household electricity bills in the short to medium term. The influx of power-intensive AI data centers has become a central election issue in some regions. A U.S. Chamber of Commerce report noted that PJM-area electricity prices surged 51% in Maryland and 41% in Illinois over the past five years. "The region has years, not decades, to deliberately make these major choices," Mills emphasized.
A policy paper released alongside Mills' letter outlined three potential pathways to bridge the "credibility gap" between attracting generation investment through high prices and protecting consumers from excessive bills. "Generators, utilities, investors, and consumers must fundamentally believe that core rules are fair, stable, and the product of a credible process," Mills wrote. Citigroup analyst Ryan Levine cautioned that PJM is taking too long to find solutions, with each proposal containing "devils in the details." He warned that prolonged indecision could lead project developers to relocate data centers elsewhere.
Earlier this year, the Trump administration called on major tech firms to self-build or independently secure power for their expanding AI data centers, urging them to sign "ratepayer protection commitments" to avoid passing costs to residential users. Companies like Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI endorsed the pledge. However, the policy lacks practical details, enforcement mechanisms, and clear timelines. Experts note that existing regulatory frameworks require utility oversight and grid standards, which cannot be bypassed by presidential statements alone. Thus, even with corporate commitments, PJM faces fundamental supply and market design challenges, likely perpetuating residential price hikes.
PJM's core issues lie in outdated grid infrastructure, slow generation growth, inadequate long-term market signals, and pressure from large-scale new loads like data centers and EV charging. Even if tech companies build their own power facilities, integration with the grid—including transmission access permits and regulatory compliance—will take time, preventing near-term resolution of supply shortages. Reports indicate PJM is exploring market reforms, long-term contracts, and investment strategies to address deficits, underscoring the critical need for grid upgrades.
The AI data center boom is unequivocally fueling a structural revaluation of global power assets. As Google, Microsoft, and Meta lead intense construction and expansion efforts, the centrality of electricity supply has magnified the investment theme that "the endgame of AI is power." If self-powering approaches become institutionalized in the U.S. and beyond, a significant portion of AI capital expenditure could shift to power equipment and grid technology. Assets with reliable generation, transmission access, long-term pricing, and data center load capacity are gaining valuation premiums akin to upstream AI supply chain bottlenecks.
PJM's warning is critical: serving 13 states and 67 million people, its current market structure cannot simultaneously ensure adequate supply and protect consumers from price spikes. Capacity prices in PJM have soared over 1,000% in two years, yet market mechanisms have failed to stimulate sufficient new generation. U.S. power shortages could emerge as early as 2027. From an engineering perspective, AI data centers represent industrial-grade baseload demand—high-density, 24/7, high-reliability, with low tolerance for interruptions.
Gold Sachs recently revised its global data center power demand forecast upward, projecting a 175% increase by 2030 from 2023 levels—equivalent to adding a top-ten global electricity consumer. The firm's strategists assert that "the end of large AI models is electricity," anticipating an unprecedented global "super demand cycle" and a "super bull market" for power stocks. The Electric Reliability Council of Texas (ERCOT) predicts peak demand could reach 367,790 megawatts by 2030, quadrupling the August 2023 record of 85,508 MW. ERCOT warns that meeting AI data center expansion and population growth may require capacity equivalent to nearly 300 new nuclear reactors by 2032.
Warnings from both PJM and ERCOT highlight that aging U.S. grid systems are ill-equipped for AI-era electricity demands, pushing tech giants into an unprecedented era of "megawatt-scale power competition" and self-supply. This data center construction frenzy is accelerating a super bull market for the most constrained segments of the power supply chain. Global capital markets are increasingly focusing on power equipment and grid infrastructure, recognizing that the AI arms race has shifted demand from GPU/TPU/server clusters to generation equipment, transformers, switchgear, transmission expansion, grid integration, and dispatch software.
In practical terms, "self-powering" typically involves behind-the-meter generation—such as large gas turbines, renewables with storage, or nuclear PPAs—coupled with dual redundancy through utility grid connections. Clear beneficiaries of this spending cycle include generators with long-term contracts, owners of nuclear/natural gas assets, providers of rapidly deployable distributed power and fuel cells, and grid technology companies serving AI data centers with large-scale capital expenditure. The strongest performers in this power stock rally will be those addressing supply bottlenecks through transmission upgrades, capacity resources, self-powering equipment, co-located energy, demand response, and high-quality PPAs/capacity contracts.