4,288 terawatt-hours. That is the U.S. Energy Information Administration’s new projection for total American electricity consumption in 2026 — up from a record 4,195 terawatt-hours in 2025, with 4,356 terawatt-hours to follow in 2027, according to the October Short-Term Energy Outlook released October 6. Three consecutive record years, each higher than the last, after roughly two decades in which national electricity demand barely grew at all.
For grid planners, this number is not trivia. It is the input to everything. Every loss-of-load-expectation study, every planning reserve margin, every integrated resource plan and capacity auction begins with a load forecast — and the arithmetic of resource adequacy only closes if that forecast is right. When the EIA moves the national demand trajectory up by nearly a hundred terawatt-hours in a single year, it quietly rewrites the adequacy math for every region in the country. This piece is about what 4,288 terawatt-hours means, what’s driving it, and what it demands of the people who keep the lights on.
What 4,288 terawatt-hours actually measures
Start with what the number is — and what it isn’t. The EIA’s figure is total U.S. electricity consumption: every kilowatt-hour used by homes, businesses, and industry over the course of the year, the sum of retail sales across all sectors. It is a measure of energy, not capacity. Planners ultimately care about the peak hour — the single gigawatt number that sizes the fleet — and energy growth does not translate one-for-one into peak growth. A data center running flat-out around the clock adds enormous energy with a steady, predictable contribution to peak; electrified heating adds energy concentrated in winter mornings and evenings. Same terawatt-hours, very different planning problems.
The trajectory matters as much as the level. The jump from 4,195 to 4,288 terawatt-hours is about 2.2 percent year-over-year growth; 4,288 to 4,356 is another 1.6 percent. Those percentages sound modest until you remember the baseline: for roughly twenty years, U.S. electricity demand grew at a fraction of a percent annually, and efficiency gains routinely offset new uses. A sustained 2 percent growth rate on a 4,200-terawatt-hour base is more than 90 terawatt-hours of new consumption a year — the annual output of roughly a dozen large power plants, materializing not gradually but in lumpy interconnection requests.
And treat the number with the respect a forecast deserves. The STEO is revised every month; weather, economic growth, and policy all move it. EIA’s October outlook explicitly ties part of the demand and price picture to extreme weather — winter storms and hotter-than-average summers. The 4,288 figure will be revised, probably more than once, before 2026 is over. But planners do not plan to the decimal point; they plan to the direction. And the direction — up, fast, and led by new kinds of load — has been consistent across every recent outlook.
What’s driving the record
The EIA names two primary drivers, and they are worth separating because they stress the grid differently. The first is data centers — specifically, facilities serving artificial intelligence and cryptocurrency, in the agency’s wording. These loads show up mainly in the commercial sector, where EIA projects 2026 sales of 1,549 billion kilowatt-hours, a new record above 2025’s 1,493 billion. A large AI data center draws power at high, flat utilization around the clock; it contributes to energy totals far more than to weather-driven peaks, but it is relentless — it does not cycle down on mild days — and it arrives in hundred-megawatt blocks through single interconnection requests.
The second driver is electrification: homes and businesses using more electricity and less fossil fuel for heating and transportation, again in EIA’s description. Residential sales are projected at 1,541 billion kilowatt-hours in 2026, also a new record (2025’s 1,515 billion was the old one). Unlike data centers, electrified load is weather-correlated — heat pumps and electric heating spike on the coldest mornings, EVs charge in the evening — which means it adds disproportionately to the peak hours that set resource adequacy requirements. Electrification doesn’t just grow the total; it changes its shape.
The telling contrast is the industrial sector: 1,055 billion kilowatt-hours projected for 2026, still below the all-time industrial record of 1,064 billion set in 2000. The old economy’s electricity peak — the mills, smelters, and factories of the late twentieth century — has never been surpassed. What has surpassed everything is the new economy’s load: digital infrastructure and electrified buildings and vehicles. The grid is not serving more of the same demand; it is serving structurally different demand, and planning models calibrated on the old load shapes are the ones now under the most strain.
Energy is not capacity
Here is the distinction that determines whether 4,288 terawatt-hours becomes a reliability problem. Resource adequacy is about having enough accredited capacity to meet peak demand with a reserve margin — the LOLE standard of 0.1 days per year. Terawatt-hours measure energy across 8,760 hours; adequacy is decided in a few dozen of them. A forecast of record energy consumption raises the odds of record peaks, but the mapping is not mechanical. What matters is coincidence: how much of the new load shows up in the same hours.
Data center load is high-coincidence in one sense and benign in another. It is always there — including at the summer evening peak — so it must be served by firm capacity, not wished away. But it is also flat and forecastable, which makes it, perversely, one of the easier loads to plan for: no weather luck involved. Electrified heating is the harder problem. It stacks onto the coldest hours, correlates across entire regions (every heat pump in the Midwest runs at once), and — as the Texas freeze demonstrated — arrives precisely when thermal generation and fuel supply are most stressed. The planners losing sleep over 4,288 are not worried about the average hour. They are worried about the coldest one.
This is where the adequacy arithmetic tightens. The U.S. system enters this demand surge with reserve margins that are already thin in places — PJM’s 2026/2027 capacity auction cleared at an 18.9 percent reserve margin, just under its 19.1 percent target, leaving essentially no slack. When the load forecast rises and accredited supply does not rise with it, the LOLE math stops closing: the expected shortfall across simulated years creeps above 0.1 days, and the region is, by definition, short. The remedies — expediting new generation, delaying retirements, raising capacity prices to pull in entry — all take years. The forecast moves in a month; the fleet moves in a decade. That mismatch is the central planning tension of the load-growth era.
The price signal is already flashing
Markets are already registering the tightness. EIA projects average U.S. wholesale electricity prices of $52 per megawatt-hour in 2026, up 11 percent from 2025, driven substantially by extreme weather — winter storms and hotter-than-average summers straining the system at the margins. Wholesale prices are the grid’s stress gauge: when energy is scarce in more hours, the average rises, and the increase lands in retail rates with a lag.
The national average hides the regional story, and planners plan regions, not nations. EIA expects wholesale prices in the PJM region — the Mid-Atlantic and Midwest, home to the densest data-center corridor on earth in Northern Virginia — to jump 41 percent, while prices in the Pacific Northwest’s Mid-Columbia hub fall 23 percent on abundant hydro. That divergence is the load-growth story in miniature: where new demand concentrates and supply can’t keep pace, prices spike; where the resource base is strong, they don’t. A single national demand number is useful context. The binding constraints are all local.
What this does to the planning workflow
Return to the adequacy workflow: forecast peak demand years ahead, run LOLE simulations, set the reserve margin, procure the capacity. The 4,288 forecast attacks the first step, and everything downstream inherits the error. Load forecasting was already the weakest link — data center load arrives in lumpy interconnection requests, not smooth trends, and weather-driven peaks are growing harder to predict. A national forecast revised upward two years running tells every regional planner the same thing: whatever peak you modeled, model it higher, and widen the uncertainty bands.
The procurement step is where the lag bites hardest. In capacity-market regions, auctions clear years ahead against demand curves built on older forecasts; if load is growing faster than the curve assumed, the cleared quantity undershoots the true need and subsequent auctions clear tighter and more volatile — the same tightness behind EIA’s projected 41 percent wholesale price jump in PJM. In IRP states, utilities must file plans that now show steeper load trajectories, which means proposing more generation, storage, and transmission — and defending larger investments before skeptical commissions. Either way, the forecast forces procurement to chase demand rather than precede it.
Meanwhile the supply side is mid-transition. EIA’s outlook has coal’s generation share sliding from 17 percent in 2025 to 16 percent in 2026 and 15 percent in 2027, natural gas holding at 40 percent before easing to 39 percent, renewables rising from about 24 percent to 27 percent, and nuclear steady at 18 percent. Read that as a planner: the energy to serve 4,288 terawatt-hours must come from a fleet that is simultaneously retiring dependable capacity and adding variable resources whose capacity credit — ELCC — is lower than nameplate and declines with penetration. Growing the numerator (demand) while re-accrediting the denominator (supply) downward is how adequacy margins evaporate without a single plant unexpectedly closing.
And then there is deliverability — the transmission question. New load concentrates where the data centers and electrified growth are; new generation builds where land and resources are cheapest. The 2,061 gigawatts sitting in U.S. interconnection queues, per Lawrence Berkeley National Laboratory’s latest accounting, is a reminder that the constraint is rarely a shortage of proposed projects — it is the wires, the permitting, and the queue processing to connect them. A terawatt-hour forecast is national; electrons move through specific congested interfaces. Every increment of demand growth raises the value of transmission that doesn’t exist yet.
The honest bottom line
A few cautions before the conclusion. First, 4,288 is a forecast, and forecasts — especially STEO vintages, revised monthly — will move. Do not anchor on the decimal. Second, energy is not capacity: the reliability question is about peaks, coincidence, and accredited megawatts, and a record consumption year does not automatically mean a record LOLE violation. Third, the EIA number is national, and as the PJM–Pacific Northwest price divergence shows, the grid experiences this growth as a set of regional crunches, not a uniform wave.
But the direction is the message. After two decades of flat demand, the U.S. power system is growing again — fast, led by loads that did not exist in the planning models of five years ago, and concentrated in ways that stress specific regions first. Every LOLE study, every reserve margin, every IRP and capacity auction conducted against the old flat-demand assumption now needs re-examination against a world adding roughly 90 terawatt-hours a year.
The planners who adapt fastest will share the traits this desk keeps coming back to: they will forecast load forward rather than backward, treat large-load interconnection as the planning input it is, and procure capacity against the peak hours that actually stress the system — not the average ones. The EIA’s number is just a number. What it demands is a planning posture built for growth, because the flat-demand era — the era every current adequacy framework was calibrated on — is over.