Loading market view
Scanning European outage signalsAcross more than 26,000 outage onsets in 10 European markets this spring, the naive link between temperature and outages turns out to be a maintenance-scheduling mirage. Isolate the forced outages, though, and a real signal appears: they stay near zero until daily-max temperature clears about 30°C, then jump roughly thirteenfold. During the late-June 2026 heatwave, unplanned outages went from a handful a day to 265 in one day.
Data source notice. Outage data is drawn from publicly available ENTSO-E Transparency Platform and ACER REMIT (UMM) disclosures, deduplicated to one record per outage event. Temperature is the daily maximum 2 m air temperature at each country's geographic centroid, taken from the Open-Meteo ERA5 reanalysis archive. Analysis window: 1 March - 28 June 2026, the period over which the platform holds dense, day-level outage coverage. Figures describe disclosed signals and observations, not findings of fault.
Ask any power trader what wrecks a summer position and "the heat" comes up fast. The intuition is that hot weather and broken power plants go together. We went looking for that link in the data — and the first answer the numbers give is a trap.
Line up every one of the 26,000-plus outage onsets disclosed across ten European markets this spring against the day's temperature, and you get a tidy positive correlation — a within-market Pearson coefficient of about 0.35. Warmer day, more outages. Case closed.
Except it isn't. That correlation is almost entirely an artefact of the calendar. Generators schedule planned maintenance for the warm, low-demand shoulder of the year, so maintenance outages pile up precisely as the weather warms — and Europe's structured outage record has also been deepening month over month. Both of those move with the season, and the season moves with temperature. Strip the slow trend out (correlate day-to-day changes rather than levels) and the relationship collapses to a Pearson coefficient of -0.02. Essentially zero. On a day-to-day basis, total outages do not follow the thermometer at all.
The two charts above share the same calendar. Temperature climbs steadily into a late-June peak. Total maintenance-driven outages, by contrast, are a noisy plateau — heavy in March, lighter in April, climbing again into the summer maintenance season, but not tracking any individual hot spell. If heat damaged the fleet, you would not see it here. The maintenance signal drowns it out.
Maintenance is scheduled — it tells you about planning calendars, not about stress. The honest test of "does heat break power plants" is the forced, or unplanned, outage: the unit that trips or derates unexpectedly. Separate those out and the picture changes completely.
Sort every market-day into temperature bands and the forced-outage rate is flat and near zero — well under a fifth of one onset per market-day — all the way up to 30°C. Then it breaks sharply: in the 30-35°C band the average market logs almost seven forced outages a day. Across all markets, days at or above 30°C carried about 6.5 forced onsets per day versus 0.5 on a typical day — a thirteenfold jump. The same pattern shows in capacity: the average market-day in the 30-35°C band saw on the order of 8 GW of generation newly declared offline, against roughly 2.4 GW in cool weather. The effect is non-linear. It is a threshold, not a slope.
The window contains one clean test. In the last week of June, a heat dome pushed centroid temperatures past 30°C across the continent, with local peaks near 38-40°C in Poland, France and Spain. Forced outages, flat for the previous three and a half months, ignited exactly in step.
Four unplanned onsets on 23 June. Eleven on the 24th. Then 154, 265 and 173 on the 25th, 26th and 27th as the heat peaked — more forced outages in three days than in the rest of the window combined. The temperature series and the forced-outage series turn the same corner on the same day.
The lift is broad, not the story of a single grid. Normalising each market against its own baseline removes any country-size effect:
Poland and the Czech Republic ran more than twenty times their normal forced-outage rate on hot days; the Netherlands and Germany seventeen to nineteen times; Italy elevenfold. Spain and France, already warm and built for it, moved least — a reminder that what stresses a fleet is heat relative to what it is engineered and acclimatised for, not the absolute number on the thermometer.
The mechanism is well understood thermodynamics, and it explains the threshold. Thermal plants — nuclear, gas, coal — are heat engines whose efficiency depends on how cold they can run the cold end. As ambient and cooling-water temperatures rise, condenser back-pressure climbs and usable output falls; past a point the unit must derate or come off. Stations on once-through river cooling face a hard regulatory ceiling: when river temperatures breach ecological limits or flows drop in a dry heat spell, output is capped by law — the recurring summer constraint on French and central-European nuclear. Solar inverters and panels lose efficiency as cell temperatures soar, and transformers and lines derate in the same conditions. All of this bites hardest at the top of the temperature range, exactly where the forced-outage curve turns up — and it arrives precisely when air-conditioning load is peaking, so supply weakens as demand surges.
The signal is real but young, and worth stress-testing as coverage grows. Three things to track:
Outage events are deduplicated from ENTSO-E TP and ACER UMM disclosures (one record per asset-and-start-time). "Forced" denotes events explicitly typed as unplanned; of the roughly 26,000 onsets in the window, about 600 are typed unplanned, 950 planned, and the large remainder are untyped — so the untyped pool may contain further forced events, meaning the true heat effect on forced capacity could be larger than measured, not smaller. Temperature uses a single national centroid as a proxy for the weather at dispersed generation sites. The dense outage record spans only four months, and the forced-outage result rests substantially on the late-June 2026 heatwave; a reporting effect coinciding with that episode cannot be entirely excluded. The headline takeaway is deliberately conservative: the raw temperature-to-outage correlation is a confound, and the defensible heat signal lives specifically in forced outages above roughly 30°C.