U.S. Grid Operator PJM to Cut Power to AI Data Centers Amid Blackout Concerns

The rapid expansion of artificial intelligence (AI) is transforming industries, but it's also putting immense strain on existing infrastructure, particularly the electrical grid. In a significant development for the tech and energy sectors, PJM Interconnection, the largest grid operator in the United States, has announced that it will implement temporary power cuts to large data centers to prevent widespread blackouts. This critical measure is set to begin as early as mid-2027 and directly addresses the surging electricity demand, largely fueled by the insatiable appetite of AI workloads. PJM Interconnection manages the electric grid across 13 states, from Illinois to the Jersey Shore, and the District of Columbia, serving approximately 65 to 67 million people. The organization’s decision underscores a growing concern about grid stability as demand for electricity skyrockets at a pace grid planners never anticipated.

The Heart of the Problem: Unprecedented Demand

For years, electricity demand in the U.S. remained relatively flat. However, recent trends, particularly the proliferation of data centers, the electrification of vehicles, and building heating systems, have dramatically altered this landscape. PJM's 2026 Long-Term Load Forecast Report confirms a trend of significant growth in electricity demand over the next two decades. According to a January 2026 report, PJM expects its summer peak load to grow by 3.6% a year, reaching about 222 GW by 2036. This represents a 10-year increase of nearly 66,000 MW. Without the impact of data centers, PJM's peak demand would actually be shrinking. The concentration of these large energy consumers is particularly acute in regions like Northern Virginia, which hosts the world's largest collection of data centers. This localized surge in electricity consumption is overwhelming existing transmission capacity and creating multi-year backlogs for new connections. Experts refer to Northern Virginia as the "epicenter" of the data industry and a hotspot for grid congestion due to this unprecedented demand shock from AI data centers.

Why Now? A Perfect Storm of Factors

Several converging factors have pushed PJM to this critical juncture, necessitating drastic measures to maintain grid reliability.

AI's Insatiable Appetite

The primary driver behind this escalating demand is the rapid growth of artificial intelligence. AI models, especially large language models and image generators, require enormous computing power, which translates directly into massive energy consumption by data centers. A conventional data center can draw as much electricity as 10,000 to 25,000 households, but a newer, AI-focused "hyperscale" data center can consume power equivalent to 100,000 homes or more. The International Energy Agency estimates that data centers consume around 1.5% of global electricity in 2024, but this demand is geographically concentrated and poised to skyrocket. By 2030, AI data centers are projected to consume up to 20% of the world's electricity. In the United States, data centers are expected to contribute nearly half of the growth in electricity demand by 2030. Some analyses suggest data centers could consume up to 34% of PJM's total power generated by 2030, with a four-fold increase in energy use by 2035. This exponential growth in AI-driven energy demand is not just about raw consumption; it's also about the consistency of that demand. Data centers require consistent, round-the-clock power at a steady level, lifting overnight and shoulder hours by a larger percentage relative to the summer afternoon peak.

Aging Infrastructure and Generator Retirements

Compounding the issue of surging demand is the state of the existing electrical infrastructure. Many of the U.S. transmission lines are nearing the end of their useful lives, with an estimated 70% of them facing this challenge. Modernizing this infrastructure requires substantial investment, with estimates suggesting $720 billion needed for grid upgrades through 2030. Furthermore, the grid is experiencing accelerated retirements of older, often fossil-fuel-based generators. These retirements are driven by unit age and environmental public policy, and they are outpacing the addition of new, mainly inverter-based resources like solar and wind. In 2025 alone, PJM received 30 deactivation notifications totaling over 6.5 GW. While renewable energy sources are being added, the pace and the intermittent nature of some of these resources create additional challenges for grid stability.

The "Bring Your Own" Mandate: PJM's Response

To address this critical imbalance, PJM Interconnection's Board issued a directive on January 16, 2026, outlining a new framework. The core of this plan is a two-tier system for large electricity consumers, particularly data centers. As early as mid-2027, data centers that fail to secure sufficient capacity in market auctions will face temporary power cuts during periods of peak demand. This initiative is formally known as "Bring Your Own New Generation" (BYONG). Under this framework, any facility adding 50 MW or more of load at a single point of interconnection must either supply its own incremental generation capacity or explicitly accept the very real risk that its power will be curtailed during grid shortages. This policy aims to shift the responsibility for new supply onto the data centers that are driving the demand, rather than burdening the general ratepayer.

Economic Fallout: Rising Costs for All

The strain on the grid and the efforts to maintain reliability have already led to significant economic consequences. Capacity auction prices have surged dramatically. For instance, the price in the 2024/2025 auction was $28.92 per megawatt-day, but for the 2026/2027 cycle, this figure exploded to $329.17 per megawatt-day—a more than tenfold increase in just two years. This increase in costs is directly passed on to consumers. Data centers are projected to pile tens of billions of dollars in additional system costs onto PJM's network through the late 2020s. Consumers served by PJM Interconnection faced an extra $9.4 billion in electricity bills in one recent summer, with another $1.4 billion increase already locked in for the following summer. Without intervention, these costs would directly impact the electricity bills of households and small businesses across PJM's footprint, potentially totaling an extra $100 billion through 2033. Adding to the complexity is the issue of speculative demand. Many companies are reserving power for data centers that are not yet built, booking more capacity than they currently need to secure their place in the interconnection queue. Utilities and grid operators must treat these requests as firm demand, leading to resource planning and upgrades for capacity that may not materialize, further exacerbating perceived shortages and driving up costs.

The Ripple Effect: What This Means for AI

The implications of PJM's decision for the artificial intelligence industry are substantial and far-reaching. Disrupted Operations: Temporary power cuts mean direct disruptions to AI operations. Businesses relying on AI for critical functions, from customer service and supply chain management to complex computational tasks and real-time analytics, could face service interruptions, lost revenue, and damage to their reputation. Location, Location, Location: This development will likely force AI companies and data center developers to reconsider their location strategies. Regions with more stable and readily available power, or those offering incentives for on-site generation, may become more attractive. The concentration of data centers in areas like Northern Virginia, while offering advantages in connectivity, now presents a significant reliability risk. Increased Investment in Self-Generation: The BYONG policy will push many large data center operators to invest heavily in their own generation capacities, such as natural gas turbines, large-scale battery storage, or even renewable energy projects. While this could enhance individual data center resilience, it adds significant capital expenditure to AI infrastructure development. Innovation in Energy Management: The necessity for greater grid flexibility will spur innovation in how data centers consume and manage power. This could include more sophisticated load-shifting technologies, intelligent energy storage solutions, and better coordination with grid operators to reduce demand during peak times without full curtailment.

• Systemic Risks of Uncoordinated Response: A recent incident on July 22, 2026, highlighted a critical systemic risk. A single transmission line outage in Northern Virginia caused multiple data centers to simultaneously switch to backup power, abruptly removing over 3 GW of load from the grid. This sudden drop in demand created a massive power surplus and caused a voltage disturbance that took PJM 10-11 minutes to stabilize, far longer than the usual milliseconds. This event demonstrated how uncoordinated responses from concentrated data centers, even when acting as designed, can destabilize the grid.

Looking Ahead: Addressing the Grid's Future

PJM's measures are part of broader efforts to secure grid reliability in the face of rapidly evolving energy demands. The grid operator has been actively working on solutions, including interconnection reforms to prioritize ready-to-proceed generator projects, and identifying baseline and network upgrades to maintain fundamental grid reliability. The Federal Energy Regulatory Commission (FERC) has also expressed "alarm bells" regarding PJM's capacity shortfalls, with the most recent capacity auction clearing nearly 7 GW below its reliability target. FERC has called meetings to discuss potential reforms, with some officials even suggesting the possibility of breaking up PJM Interconnection into smaller, more manageable pieces if reforms are insufficient. Proposed solutions extend beyond curtailment and self-generation. Experts advocate for greater investment in grid-enhancing technologies (GETs) and large-scale energy storage, such as batteries, which can provide crucial flexibility and reliability, especially with the integration of more intermittent renewable sources. PJM has also initiated a "Reliability Backstop Procurement" to attract more generation to the grid and facilitate private bilateral contracts to address near-term reliability needs. The challenge is immense. The future of AI's growth is inextricably linked to the stability and capacity of our electrical grids. As AI continues its explosive trajectory, the energy infrastructure must evolve rapidly to prevent a bottleneck that could hinder technological progress and economic development.

Frequently Asked Questions

What is PJM Interconnection?

PJM Interconnection is the largest independent system operator (ISO) in the United States, managing the wholesale electricity market and operating the transmission grid in all or parts of 13 states and the District of Columbia. It serves approximately 65 to 67 million people.

Why is PJM planning power cuts for data centers?

PJM is planning power cuts for large data centers to prevent blackouts due to an unprecedented surge in electricity demand, primarily driven by energy-intensive AI data centers. This demand is outpacing the addition of new generation capacity and straining the existing, aging grid infrastructure.

When will these power cuts begin?

PJM Interconnection has indicated that temporary power cuts for data centers could begin as early as mid-2027.

What is the "Bring Your Own New Generation" (BYONG) policy?

The BYONG policy is a framework introduced by PJM where large data centers (adding 50 MW or more of load) must either provide their own generation capacity or accept the risk of having their power curtailed during periods of grid stress and potential blackouts.