The 175 GW Crisis: America's Power Grid Cannot Keep Up with AI Data Centers

Jan 21, 2026 Written By Blake Crosley

The United States faces a 175 gigawatt capacity shortfall by 2033—equivalent to the electricity consumption of 130 million homes—as AI data centers strain an electrical grid where 70% of infrastructure approaches the end of its operational life.¹ Data centers have already added $23.1 billion to power procurement costs in the nation's largest grid over three consecutive auctions, with technology companies now accounting for nearly half of all capacity market expenses.² Federal agencies announced accelerated interconnection plans on January 20, 2026, attempting to compress decade-long timelines even as utilities struggle to build generation capacity fast enough to meet demand projections that have tripled in less than two years.

TL;DR

The U.S. power grid confronts its most severe reliability challenge in decades. Schneider Electric forecasts that peak electricity supply will fall short of demand by 2028, with the gap reaching 175 GW by 2033. Data centers drove 40% of PJM Interconnection's $16.4 billion capacity auction in December 2025, adding $6.5 billion in costs that flow through to consumer electricity bills. The Department of Energy has directed FERC to finalize rules by April 2026 allowing faster interconnection of large loads, while Bloomberg reports a new federal plan to accelerate data center grid connections. Meanwhile, utilities have placed orders for 32 GW of large gas turbines since 2024—a 20-year high—but new generation takes years to build while data center demand grows monthly.

The Numbers Behind the Crisis

Grid Strategies found that utility forecasts for five-year peak demand growth jumped from 38 GW in 2023 to 128 GW in 2024—a 237% increase in projected need within a single year.³ Data centers account for most of this revision.

US data center power demand trajectory:

*BloombergNEF revised its 2035 forecast upward by 36% in late 2025, while other analysts project even higher figures.⁴

The supply side cannot match this trajectory. New generation sources—whether solar farms, gas plants, or nuclear facilities—require 7-12 years from initial planning to delivering electrons. Data centers can go from groundbreaking to operational in 18-24 months. This fundamental mismatch creates the capacity gap.

The $23 Billion Bill

PJM Interconnection serves nearly 65 million people across 13 states and the District of Columbia, operating the largest competitive wholesale electricity market in the world. Its capacity auctions reveal what data centers cost the grid.

PJM capacity auction results (data center impact):

Capacity prices have soared from $28.92 per megawatt-day in 2024/25 to $333.44/MW-day in 2026/27—an eleven-fold increase.⁵ These costs flow through to electricity bills. In Washington D.C., Pepco residential customers saw bills rise by $21 per month starting in June 2025, with approximately half attributable to capacity market price increases driven by data center demand.⁶

The Institute for Energy Economics and Financial Analysis warned that "projected data center growth spurs PJM capacity prices by factor of 10," raising questions about whether the current market structure can sustain exponential load growth without fundamentally restructuring cost allocation.⁷

Aging Infrastructure Meets Unprecedented Demand

The grid straining under this load was not designed for it. Most U.S. electrical infrastructure was built between the 1950s and 1970s. As of 2023, 70% of transmission lines and transformers had exceeded 25 years of operation.⁸

Clift Pompee, VP of power and emissions at Compass Datacenters, described the situation: "2026 is a pivotal year for the future of the U.S. power grid. Unprecedented load growth is exposing the aging nature of our grid."⁹

Regional stress points:

PJM Interconnection: The nation's largest grid operator projects a 6 GW shortfall against reliability requirements by 2027.¹⁰ December 2025's capacity auction fell 6,623 MW short of its reliability target, with data centers responsible for nearly 5,100 MW of the demand surge.

ERCOT (Texas): Interconnection requests have exploded from 56 GW a year ago to over 205 GW currently—nearly quadrupling in twelve months. Over 70% of these requests come from data centers, with another 10% from cryptocurrency mining.¹¹ Reserve margins could fall into risky territory after 2028.

NERC Assessment: The North American Electric Reliability Corporation warned of "elevated risk" of summer electricity shortfalls in 2026 and beyond across all three interconnection regions.¹²

The Edison Electric Institute reports its members spent $1.3 trillion over the past decade enhancing grid infrastructure, with another $1.1 trillion planned over the next five years.¹³ These investments address existing maintenance backlogs—they do not account for the additional capacity data centers require.

Federal Response: Accelerating Interconnection

Federal agencies have moved to address the bottleneck where new load meets aging infrastructure: the interconnection process.

January 20, 2026 - Bloomberg reports federal grid plan: The Department of Energy announced a new initiative to accelerate data center connections to the grid, responding to industry complaints that interconnection delays threaten U.S. competitiveness in AI development.¹⁴

December 18, 2025 - FERC orders PJM overhaul: The Federal Energy Regulatory Commission found PJM's existing tariff "unjust and unreasonable" and ordered the grid operator to establish clear rules for data center colocation at power plants. The ruling creates pathways for facilities to connect directly to generation without waiting for transmission buildouts.¹⁵

DOE directive to FERC: Secretary of Energy Chris Wright directed FERC to finalize rulemaking by April 30, 2026, allowing customers to file joint co-located load and generation interconnection requests. The proposed rules would significantly reduce study times and grid upgrade costs.¹⁶

Key regulatory deadlines:

The reforms aim to address a fundamental inefficiency: data centers currently wait 8-12 years in some interconnection queues, during which time technology generations turn over multiple times. A facility conceived when H100 GPUs were state-of-the-art might finally connect to the grid when H300 or successor chips have made original designs obsolete.

The Generation Gap

Faster interconnection processes do not create electrons. The physical infrastructure to generate electricity requires years to build regardless of permitting efficiency.

Generation orders and timelines:

Large gas turbine orders hit a 20-year high of 14 GW in 2024, driven by data center load growth. This trend accelerated into 2025, with 18 GW of U.S.-bound gas turbine orders placed in the first half of the year alone.¹⁷

However, timelines for new generation remain measured in years:

Data centers face pressure to "bring their own generation." Bloom Energy's 2026 Power Report found data centers planning to reduce grid reliance through on-site generation, including fuel cells, natural gas turbines, and solar-plus-storage installations.¹⁸

This shift toward distributed generation represents both solution and complication. On-site power reduces grid strain but also reduces utilities' ability to spread infrastructure costs across larger customer bases, potentially raising rates for residential users who cannot self-generate.

Industry Adaptation Strategies

Organizations deploying AI infrastructure have developed several strategies to navigate the capacity crisis.

Co-location at power plants:

FERC's December 2025 order explicitly enables this approach. Data centers connect directly to generation facilities, drawing primary power on-site while maintaining grid backup. Amazon's deals with Talen Energy for nuclear power exemplify the model.¹⁹

Benefits:

• Bypasses transmission queue entirely
• Accesses power at generation cost rather than retail rates
• Provides dedicated capacity without competing for grid resources

Challenges:

• Limited to locations near suitable generation
• Raises questions about generator obligations to serve grid during shortages
• May require capital contribution to generation capacity

Vertical integration:

Google's $4.75 billion acquisition of Intersect Power in December 2025 represents the most aggressive form of this strategy—hyperscalers directly owning generation assets rather than purchasing power through contracts.²⁰

Benefits:

• Full control over development timeline
• Eliminates dependency on third-party developers
• Enables "energy park" co-location from initial design

Challenges:

• Massive capital requirements
• Transforms technology companies into power companies
• May face regulatory scrutiny as market consolidation

Geographic diversification:

Some operators are shifting development to regions with available capacity, even if those locations are suboptimal on other factors like fiber connectivity or labor markets.

Midwest and Southeast markets have attracted data center investment as traditional coastal markets reach capacity constraints. Wisconsin's experience with Microsoft demonstrates both the opportunity and the community resistance such expansion encounters.

Demand flexibility:

Emerging approaches include designing data centers that can reduce consumption during grid stress events, providing "demand response" services that utilities compensate.

AI training workloads, while power-intensive, can often pause and resume without significant penalty. Inference workloads serving real-time applications have less flexibility. Facilities architected to separate these workload types can offer meaningful demand response while maintaining service levels on critical functions.

The 2026-2028 Crunch

Multiple analyses converge on 2026-2028 as the most severe period for grid stress. Supply additions lag demand growth, with the gap narrowing only as delayed generation projects finally reach completion.

Critical watch points:

Summer 2026: NERC's elevated risk assessment suggests potential shortfalls during peak demand periods. Grid operators may implement emergency protocols more frequently.

June 2026: PJM's 2028/2029 capacity auction will price power for the delivery year when shortfalls are projected to peak. Prices could set new records.

2027: PJM reliability requirements fall 6 GW short under current projections. Emergency imports from neighboring regions may be required.

2028-2029: If gas turbine orders from 2024-2025 complete on schedule, supply begins catching up to demand. But any construction delays extend the crunch period.

Infrastructure operators should plan for:

• Higher electricity costs flowing through from capacity markets
• Potential reliability events requiring backup power activation
• Regulatory changes affecting interconnection and cost allocation
• Community and political pressure on data center expansion

The Infrastructure Deployment Perspective

The capacity crisis creates specific requirements for organizations planning AI infrastructure deployments.

Power strategy becomes primary site selection factor:

Traditional data center site selection prioritized fiber connectivity, labor availability, and real estate costs. Power availability—both current capacity and future expansion potential—now dominates evaluation criteria. A site with excellent connectivity but constrained power is less valuable than a site with adequate connectivity and abundant power.

Backup power requirements intensify:

Facilities in reliability-stressed regions need more robust backup systems than historical designs assumed. Extended grid outages become more likely during the 2026-2028 crunch period. Generator fuel supply contracts require scrutiny—regional shortages during widespread outages could leave facilities stranded.

Cost modeling must account for capacity market dynamics:

Electricity cost projections based on historical rates will prove inaccurate. Capacity prices have increased eleven-fold in PJM alone. Organizations should model scenarios with continued capacity price escalation rather than assuming mean reversion.

Community relations become operational requirement:

The political backlash against data centers—manifesting in everything from Wisconsin's regulatory debate to bipartisan Congressional concern—affects permit timelines, utility rate structures, and long-term operational stability. Organizations that treat community engagement as optional face increasing risk of project delays or cancellations.

What the Federal Response Means

The federal push to accelerate interconnection addresses one bottleneck but not the underlying constraint. Faster processing of queue applications does not create generation capacity, build transmission lines, or upgrade aging infrastructure.

The reforms do enable:

• Faster access to available capacity where it exists
• Direct data center connections to co-located generation
• Reduced study costs for interconnection applications
• Clearer rules for hybrid configurations

They do not address:

• Total generation capacity relative to demand
• Transmission constraints between regions
• The cost of new infrastructure and who pays
• Grid reliability during the construction period

Secretary Wright's directive positions AI infrastructure as a national priority warranting expedited treatment. This framing may support continued federal intervention if initial reforms prove insufficient. However, electrons cannot be regulated into existence—physical infrastructure must be built.

Looking Ahead

The 175 GW shortfall projection assumes current demand trajectories continue. That assumption could prove conservative if AI capabilities and applications expand faster than anticipated, or optimistic if economic factors slow data center construction.

What appears certain: the U.S. power grid faces the most significant reliability and capacity challenge since its construction seven decades ago. The infrastructure that powered postwar industrial expansion cannot power the AI revolution without fundamental transformation.

Organizations planning AI deployments should build power strategy into their technology strategy from initial conception. The companies that secure power access through direct ownership, long-term contracts, or co-location relationships will have capacity to operate. Those that assume power will be available when needed may find themselves capacity-constrained regardless of how many GPUs they can purchase.

The physical infrastructure requirements span generation, transmission, distribution, and on-site power systems—each with specialized engineering demands. The next three years will determine whether the grid transformation matches the AI transformation, or whether power becomes the limiting factor for American technological advancement.

References