Nuclear refuels in spring and fall and runs flat through the peaks, so the only honest comparison
is the same season in a different year. Pick one market to compare it across years, or several to compare them
against each other.
Fall 2026 averaged 1,527 MW off line across
PJM, 4.7% of installed nuclear capacity, peaking at
7,216 MW on October 13.
5 refueling outages started in the window,
median 18 days.
Only 107 days of the season have been reported so far, so treat the
comparison as partial.
Every line is the same calendar window in a different year, so the curves are directly comparable. The current
season is drawn heaviest. Where the lines separate is where a year was genuinely different, not where the
calendar moved.
Add 53 more seasons to the comparisonIndividual · $165/mo
Your plan reaches back to the current season. The record runs to Spring 2000. Most readers do not want to click through the archive a season at a time. The depth is worth having in bulk, to fit a model against, which is what the API and the bulk export are for.
Spring seasons in cyan, fall in gold. Fall is normally the heavier of the two: more units
refuel into the winter peak than out of it.
The hatched bars are seasons outside your history window. They are drawn at a fixed height to
show the record exists, not at their real value, so the chart never hints at a figure you have not unlocked.
Unlock the full record.
Fall 2026 by Market
heaviest first · click a market to add it to the comparison
How to read older seasons. The megawatt columns are built from the
daily power record of the units operating today, carried back; the outage counts include units that have
since shut down. Across recent seasons those are the same fleet, so the two agree. Go back far enough and
the megawatt figures describe today's fleet in an older year, which is the right basis for comparing supply
but not for counting what the industry did at the time.
Season history follows the same backward window as the national curve and the ISO
breakdown, so there is one rule across the product rather than a special case here. The full record runs to
Spring 2000; for model fitting, take it through the
API or the bulk export rather than a season at a time.
The record says what happened. What about what is coming?Individual · $165/mo
The forward curve is the same question pointed the other way: a probability band on MW offline out to 18 months, with every published outage date and every projected one behind it.
