EIA's November outlook sees U.S. power generation at 4,390 billion kWh in 2026: how much is AI?

The U.S. government's energy forecasters published a fresh outlook on November 10, 2026, and several outlets read it as an AI story. The numbers deserve a more careful look than the headlines. The Energy Information Administration's summary describes steady, ordinary-looking growth in power generation, and it explains that growth in terms of commercial demand and weather. Whether artificial intelligence is the hidden driver depends on what you can and cannot read into those words.

What EIA actually forecasts

In its November Short-Term Energy Outlook, with inputs finalized on November 5, EIA says it expects the U.S. electric power sector to generate almost 4,390 billion kilowatthours in 2026, up 2.7% from 2025. It then expects growth to slow to 1.2% in 2027. EIA says 2026's rise is "driven by rising demand from the commercial sector and increased cooling needs from warmer-than-average summer temperatures," while milder weather is forecast to reduce residential demand in 2027.

On the supply side, EIA says most of the generation growth over the forecast period comes from utility-scale solar plants. Its overview table has solar rising from 7% of U.S. generation in 2025 to 8% in 2026 and 10% in 2027. Natural gas goes from 40% in 2025 and 2026 to 39% in 2027, and coal holds at 15% in 2026 and 2027. EIA also expects U.S. coal consumption to fall 7% to 419 million short tons in 2026. Nuclear stays at 18%.

Where AI enters the story

The EIA summary page we reviewed does not mention data centers or AI. The connection arrives through the surrounding coverage. A roundup by The Neuron on November 11 cites a Reuters report that the agency expects record electricity use in 2026 and 2027, partly from AI data centers, with consumption rising from 4,216 billion kWh in 2025 to 4,370 billion kWh in 2027. We have not been able to open the Reuters piece or the EIA table behind those consumption figures, so treat them as reported, not confirmed by us. They are also a different measure from the generation figure above: consumption and generation are not the same line in EIA's accounting.

That is the first discipline. A power forecast that says "commercial sector" is a statement about offices, retail, hospitals, schools and server farms together. Data centers sit inside that sector, but the headline category is much larger than AI.

What an independent academic review says

The 2026 AI100 report from Stanford, released the same week, takes a stronger view of the physical footprint. Its key takeaways say the headlong rush to build data centers "is straining current energy and freshwater supplies, driving up global CO2 emissions, and reshaping national energy policies," and that there is a growing backlash against data center construction in several countries, including the United States. That is a qualitative judgment from an academic panel rather than a forecast of U.S. kilowatt-hours, and it fits what we have seen at the ballot box in early November, when voters in several places weighed limits on new data centers.

Put the two together and you get a defensible reading. EIA's published summary shows national demand growing at a moderate pace. AI100 and the local politics show the growth is concentrated and contested. National averages can look calm while a handful of utility territories absorb huge loads.

Why the national number can mislead

A 2.7% rise in national generation says little about where the new load lands. A gigawatt-scale campus in one county can strain local transmission and capacity even if the national total moves by a fraction of a percent. Our site-by-site AI data center power tracker follows those campuses, and some builders are bypassing the grid altogether with on-site generation. Local utilities, state regulators and county boards, not national forecasters, decide whether each project gets power and on what terms. A national forecast cannot see any of that.

Cooling deserves a second look as well. EIA names warmer-than-average summer temperatures as a driver, so a hot year can add to 2026 demand without any change in computing. That makes year-over-year comparisons noisy, and it is one more reason not to credit the whole increase to AI.

The other discipline is date. EIA forecasts are revised monthly and the agency stresses that its November model does not account for events after November 5. The market shock of that week, covered in our report on the November 5 AI stock selloff, falls just outside that window. If financing for data centers tightens, the 2027 load could move.

What we would watch

Look at three things in the EIA's detailed tables rather than the summary: commercial-sector sales growth by month, any language about data centers in the electricity section, and the regional breakdown. Those tell you whether AI is a rounding error nationally or the dominant driver in a few regions.

Also check how the 2027 slowdown to 1.2% is explained. EIA ties it to milder weather and residential demand, not to any pause in computing buildout. If load from large computing campuses really were the main force, the forecast would be less sensitive to weather.

Our position: do not repeat "AI is driving record U.S. power demand" as a fact from this release. The honest statement is narrower. EIA expects generation of almost 4,390 billion kWh in 2026 and credits commercial demand and cooling; independent academic work says data center construction is straining energy supplies; and the share that is AI is still not something this forecast lets you state to the decimal. The December 8 release is the next checkpoint.