A year ago, grid operators across the country were bracing for the worst. The U.S. Energy Information Administration had just published its most aggressive electricity demand forecast since 2000, driven largely by a wave of AI data centers racing to connect to the power system. Utilities in Virginia, Texas, and the Midwest warned that reserve margins could thin to dangerous levels during peak summer heat. Some analysts openly discussed the possibility of rolling blackouts.
Now, heading into summer 2026, the picture looks meaningfully different. Much of that new data-center load is arriving later than anyone expected. Permitting delays, transmission study backlogs, and equipment bottlenecks have pushed the sharpest demand spikes past 2027 in several key regions, buying grid planners extra time they did not think they had. The blackout threat has not vanished, but it has loosened its grip on the near term.
The demand surge is real, but the timeline has shifted
The EIA’s forecast was not wrong about direction. U.S. electricity consumption is climbing, and data centers are a primary reason. Hyperscale facilities operated by Microsoft, Amazon, and Google require enormous, continuous power draws, often hundreds of megawatts per campus, and dozens of new sites are under construction or in advanced planning stages.
But a regional breakdown of the EIA’s own data on fossil generation trends reveals a critical nuance: in several of the country’s busiest data-center corridors, the bulk of expected load growth from hyperscale computing lands after 2027, not during the 2025-to-2027 window that had grid operators most worried. That distinction matters because power plants, substations, and high-voltage transmission lines take years to build. If demand arrives before new supply is ready, the math turns dangerous. The post-2027 shift gives planners additional runway to bring capacity online before the heaviest loads materialize.
In PJM Interconnection, the regional grid operator covering 13 states from Virginia to Illinois and home to the largest concentration of data centers on Earth in Northern Virginia, the interconnection queue has swelled to hundreds of gigawatts of proposed projects, both generation and load. But moving through that queue takes time. Transmission studies required before a new facility can draw power often stretch longer than the construction of the building itself, creating a bottleneck that has effectively throttled how fast data centers can energize.
Why the hookups are lagging
The slower pace of data-center connections is not a sign that tech companies are pulling back. Capital spending on AI infrastructure remains enormous. What has lagged is the final step: completing the grid studies, securing the permits, and flipping the switch to draw power at full capacity.
Several forces are at work. Local permitting timelines have stretched in communities where residents push back against the noise, water consumption, and sheer footprint of large data campuses. In Loudoun County, Virginia, the epicenter of U.S. data-center development, zoning debates have grown more contentious as facilities expand into residential buffer zones. In parts of central Texas and central Ohio, similar friction has slowed approvals.
Transmission study backlogs compound the problem. Regional grid operators like PJM, ERCOT, and MISO process interconnection requests on a first-come, first-served basis. Earlier projects must be studied before newer applications can advance, and the sheer volume of filings has created a years-long queue in some regions. Equipment lead times for large power transformers and switchgear, many of which are manufactured overseas, have also extended since the pandemic-era supply chain disruptions.
The Federal Energy Regulatory Commission clearly views the bottleneck as a systemic issue, not just a regional inconvenience. FERC opened a formal rulemaking in early 2026 examining how large electricity consumers, including data centers and advanced manufacturing plants, connect to the interstate transmission system. The proceeding is exploring whether to standardize study processes, clarify who pays for grid upgrades, and give transmission providers clearer authority to manage clusters of similar projects. Final rules have not yet been issued, leaving developers and utilities in a holding pattern as they try to anticipate how the framework might change.
The supply side is catching up, but unevenly
The breathing room created by delayed data-center loads has coincided with a burst of new generation coming online. Utilities across the Southeast and Midwest have accelerated gas-fired projects and life extensions for existing plants. Solar and battery storage installations continue to set annual records, adding flexible capacity that can help cover peak afternoon demand. And high-profile nuclear developments, including the planned restart of the Palisades plant in Michigan and Constellation Energy’s agreement to bring a unit at Three Mile Island back into service, signal a broader willingness to lean on firm, carbon-free power to meet AI-era loads.
But the buildout is uneven. Regions with the densest data-center pipelines do not always have the most new generation in the queue. Northern Virginia, for instance, sits on a transmission-constrained corridor where adding large amounts of new supply requires costly upgrades to lines that already run near capacity. Texas has more room to build, but ERCOT’s energy-only market design has historically struggled to incentivize the kind of firm, always-available generation that data centers need most.
Natural gas storage levels offer an indirect but useful signal. Weekly EIA storage reports heading into summer 2026 show inventories that are more comfortable than many analysts projected. Slower data-center energization is one plausible contributor to that cushion, because gas-fired power plants are the marginal supply source most sensitive to sudden demand shifts. But storage levels also reflect weather patterns, pipeline flows, and liquefied natural gas export volumes, so isolating the data-center effect requires more detailed modeling than is currently public.
What is still missing from the picture
The strongest evidence in this story comes from federal primary sources. The EIA’s demand forecast and its regional fossil generation analysis are built on utility-reported data, interconnection queue filings, and econometric modeling. These are not opinion surveys. They represent the federal government’s best estimate of where power demand is headed.
What the available evidence does not yet provide is direct, on-the-record testimony from data-center operators about why their grid connections have slipped. Interconnection queue data from PJM, ERCOT, and MISO would show exact withdrawal rates, study completion timelines, and capacity reservations. That data exists but is scattered across individual grid operators and has not been aggregated into a single federal dataset covering the full picture. Until it is, the precise cause of the delay, whether permitting friction, equipment shortages, financing constraints, or strategic timing by operators, remains partly opaque.
Another gap: how much flexible load data centers can realistically offer during emergencies. Some operators have suggested they could curtail noncritical computing tasks during extreme peaks, effectively acting as demand-response resources. But those claims have not been tested at scale during a real heat wave or polar vortex, and regulators have not yet incorporated such flexibility into their reliability planning assumptions in a meaningful way.
A window that will not stay open forever
For households and businesses heading into another hot summer, the practical takeaway is straightforward: the immediate blackout risk that dominated headlines through 2025 has eased. Grid reserve margins in most regions look healthier than worst-case projections suggested, thanks to the combination of delayed large loads and incremental additions of new generation and storage.
But the load is not gone. It is deferred. If data-center projects that stalled in permitting or interconnection queues clear those hurdles in a compressed window after 2027, the grid could face a steeper ramp than a gradual buildup would have produced. That scenario would test whether the investments made during this breathing room were large enough and fast enough, and it would expose any lingering weaknesses in regional coordination, especially where multiple large campuses cluster on the same transmission corridors.
Two indicators are worth watching in the months ahead. First, whether FERC’s large-load interconnection proceeding produces new rules that speed up or slow down the queue for big consumers. Faster, more predictable processes could pull demand forward; more stringent requirements or higher upgrade costs could extend the current period of deferral. Second, how often regional grid operators invoke emergency measures during heat waves this summer. A decline in such events, despite rising baseline load, would suggest the system is adapting effectively.
The extra time on the clock is real. Whether it gets used to build a grid that can handle what is coming, or gets wasted while the same collision of growth and constraints gathers speed, is a decision that utilities, regulators, and policymakers are making right now. The evidence says the risk has not disappeared. It has simply agreed to arrive a little later.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
