The story of one hot evening. PSE's third capacity alert, yet balancing energy stayed cheap

Tension in the system, like panic on the market, does not appear suddenly. It builds up. It was the end of June and a broad "heat dome" slid over Europe — a high-pressure system that day after day pushed temperatures 14–18°C above normal and gave Western Europe the hottest June in the history of measurements (+3.05°C against the 1991–2020 norm). On 28 June records fell on both banks of the Oder: 41.7°C in the German town of Coschen (Germany's all-time record) and, opposite it in Słubice, 40.5°C (the highest temperature in the history of Polish measurements, IMGW-PIB). Such weather is a double blow for the system. On one hand, millions of air conditioners run at full power and do not switch off after dusk, because hot nights bring no relief. On the other hand, the same heat dome that heats also puts the wind to sleep. In the week ending 28 June, Germany recorded the lowest wind generation of the whole year. In Poland it was identical: wind output fell to 203 MW on 28 June, against more than a dozen gigawatts of installed capacity. At the very moment when cooling was working most intensively, one renewable source was disappearing from the balance almost entirely.

At the same time French nuclear power was dropping out of European supply. The heat warmed the rivers cooling the reactors (the Seine at the Nogent plant, the Rhône at Bugey, the Garonne at Golfech) above the thresholds at which the law requires limiting the discharge of warm water. EDF shut down a total of around 6.2 GW of nuclear capacity, close to 10 percent of its fleet. To convey the scale: that is more than the capacity of the Bełchatów power plant, the largest in Poland — the equivalent of roughly six large nuclear units switched off at once. The effect was immediate across Europe. French exports slid from 10–12 GW to around 3 GW, French spot prices reached their highest level since January 2025, and at the peak the continent recorded levels unseen since the 2022 crisis (in Belgium on 24 June the price briefly broke through 1,000 €/MWh). For Poland this meant one thing: the cheap imports that normally cushion the evening peak were vanishing. The buffer the system usually counts on after sunset was closing, and this was to hit hardest on 30 June, when the evening cross-border exchange balance fell almost to zero.

The entire mechanics of these three days fits into a single curve: residual load — that is, demand less photovoltaic and wind generation. It is the capacity that the rest of the system must cover: thermal units, pumped-storage plants, batteries and imports. When the sun shines, residual load falls almost to zero. When it goes out, everything else has to step in for it in a flash. At midday, photovoltaics flooded the market. On 28 June (Sunday) PV reached 12 GW, residual load collapsed to 4,830 MW at 10:30, and PSE had to curtail photovoltaics — momentarily by as much as 2,475 MW, over 44 quarter-hours of the day (June 2026 was in fact a record: 286.6 GWh of renewable energy was curtailed in total). The excess was so large that at midday Poland was exporting power, and prices on the balancing market fell deep below zero, to −2,484 PLN/MWh at 13:30. Then came the solar cliff. When photovoltaics went out after 18:00 and the wind did not wake up, residual load shot up: from a midday minimum of 4,830 MW to 17,562 MW in the evening. In the four-hour window of the evening ramp alone the increase reached 8.6 GW. The same sequence is visible on each of the three days (a ramp of 6.7–8.6 GW/4 h). This is no longer the "evening peak" known from decades past. It is a wall the system has to build in a few hours.

Each of the three days responded differently, and the differences say more than the records themselves. 28 June (Sunday): the day-ahead market calm, the balancing market wild. Low weekend demand and peak PV gave a mild day-ahead index (averaging 380 PLN/MWh), but the balancing market went crazy in both directions: from −2,484 PLN at midday to +3,215 PLN in the evening. All the tension came down onto balancing — a day in which it was not the exchange but the real system that took the hit. 29 June (Monday): "expensive together with Europe". The return of full demand from factories and offices (peak 22,858 MW at 18:30), combined with an expensive continent, drove both indices up together: in the evening the day-ahead index was 1,229 PLN and the balancing energy price (CEB) 1,341 PLN, without a single negative quarter-hour in the whole day. When the exchange and the balancing market move in step and high, it is a sign of a fundamental deficit, not a local one.

The morning of 30 June. Photovoltaics were only just ramping up, but the operator was already looking not at what was, but at what the evening would bring: rising, heat-boosted demand, a sun that in a few hours would begin to fade, weak forecast wind and closed imports. At 09:51 PSE issued a communiqué. For three evening hours — 18–19, 19–20 and 20–21 — it announced capacity-alert periods on the capacity market: a formal call on capacity-market units to fulfil the capacity obligation for which they are paid. Generators are to make capacity available or feed it into the grid, and demand-side response (DSR) units are to reduce their offtake. And then the situation deteriorated before one's eyes. Wind generation crumbled throughout the morning: from 1,086 MW at six o'clock to just 187 MW at eleven, the lowest level of the whole day, exactly in the hours when the operator was issuing and extending its communiqué. Less than three hours after the first alarm, at 12:43, PSE issued a second communiqué and added a fourth hour, 21–22. It gave the reason plainly: "The need to announce capacity-alert periods stems, among other things, from high demand associated with the heat, low forecast wind generation and the unavailability of capacity in conventional units." Reduced capacity was then being given up by the Kozienice, Turów, Ostrołęka and Żerań units: individually minor losses, but together enough to deepen the reserve shortfall. The capacity alert is a rarely used remedial measure of the capacity market, triggered when the forecast capacity surplus falls below a safe level; the measure of last resort is only the power-restriction levels. Previously PSE had announced capacity-alert periods only twice: on 23 September 2022 (two hourly windows) and on 6 November 2024 (three windows). 30 June 2026 was the third such day in the history of the capacity market, and at the same time the broadest, with four windows, and the first triggered by heat. The market did not need a communiqué to sense the danger. Already the day before, at the day-ahead auction for delivery on 30 June, the evening price soared: in the 18–22 window it averaged 1,578 PLN/MWh, and at the peak at 20:30 it touched 2,290 PLN/MWh — more than four times the normal daily average. Everything pointed to a crisis.

And then the evening came and it turned out that the price of balancing energy in those same critical hours averaged just 374 PLN/MWh. The balancing energy price (CEB) shows the settlement price of balancing energy once the outturn is known, and it is not a measure of the total cost of securing the system, but it does tell how expensive the current closing of the balance was. It was a level lower than in an average working evening of the same month (around 900 PLN/MWh). The spread between the day-ahead and the balancing market reached almost 1,800 PLN/MWh at the peak. The day before, the market had priced in panic. Reality turned out to be calm. Three things came together, and all of them worked on the system's side. First, the alarm concerned the forecast, not reality. The operating reserve forecast the day before delivery collapsed in the evening to just 1.2 GW at 20:15, below the required level. It was this trough that triggered the capacity alert. But in the outturn the operating reserve held ~3.4 GW, and the surplus in the wider KSE capacity balance reached 7.6 GW; the margin the market feared, for the most part, simply appeared. Second, the wind woke up exactly as the sun was going out. As photovoltaics collapsed in two hours from 4,250 MW (18:00) to 479 MW (20:30), wind generation was rising: from 869 through 1,265 (20:00) to 1,880 MW (22:00). The dead minimum of wind fell in the middle of the day, not at the peak. Two sources that in a hot evening can fail at once this time shook hands. Third, the dispatchable units carried out the ramp without running out of breath, increasing output by 5,333 MW between 14:00 and 20:00. At the evening peak the realised operating reserve stood at around 3.4 GW, and the surplus of available capacity in the wider KSE balance reached around 7.6 GW. The system did not come close to the edge for even a moment.

Here is the crux, and the point the market often misses. The day-ahead index and the balancing energy price measure two different things. The day-ahead is an ex-ante price, shaped by participants' expectations about the balance and by the pricing of the risk of a difficult evening; an evening deficit with zero photovoltaics, weak wind and no imports was fully probable D-1, so the market bought insurance. The balancing energy price is an ex-post settlement price, set already with knowledge of the actual system run, once uncertainty had settled on the favourable side. It is worth adding a nuance: the marginal aFRR activations were expensive — the upward activation price jumped above 1,700 PLN/MWh in the evening — but that was a thin slice at the top of the ramp. The bulk of the energy was delivered by scheduled units at normal cost. A high margin, a cheap average — that is the signature of a well-secured system.

What happened on 30 June is a pattern perfectly familiar from financial markets: "buy the rumour, sell the fact". The day-ahead market priced the evening on the rumour, on the fundamentals: the heat, the dead wind, the fading French nuclear fleet and the closed imports. It priced them, moreover, earlier than the operator itself, because the day-ahead auction closed on the afternoon of 29 June, before PSE published the reserve forecast falling to 1.2 GW and before it announced the capacity alert in the morning. And when the facts arrived — the realised reserve, milder weather and returning wind — the price of balancing energy settled low. These three days were not a harvest for storage; in raw numbers June was better. But they showed something more important: where the future advantage lies. The evening spread was enormous, but it does not collect itself. It will be captured not by whoever has the largest storage asset, but by whoever forecasts best and decides fastest. And there will only be more such evenings, because as the share of renewables grows volatility rises, and the example of France shows that even a nuclear fleet is not immune to the weather. Capacity alone ceases to be enough. What counts is the forecast, the decision and readiness on both sides of the meter. Someone else was watching over all of this. In the strategy "The Guardian and the Architect. PSE's Strategy to 2040", published on 16 December 2025, the operator explicitly took on two roles: the architect, who rebuilds the grid for a zero-emission mix, and the guardian, whose task is the safe operation of the system here and now. On 30 June we saw that guardian in action. PSE read the narrow reserve margin and reached for a tool used only twice before. The forecast reserve fell short of the required level by little, not by several gigawatts, so it was a pinpoint intervention, not a rescue. And here is the crux. A capacity alert is not measured by how much capacity was actually activated, but by how much stood ready in case the evening went the other way. This time the market helped, the weather relented, the wind returned and the real deficit turned out to be small, so the guardian reached for little. But had the wind not woken up, had demand not come down from the forecast, had another unit dropped out — behind the alert stood the whole pool of contracted capacity, ready to enter the game. The value was not the consumption, but readiness for a scenario that did not materialise. The security of the system is measured not by how much capacity was activated, but by how much stood in reserve when the alarm sounded.