The Night France Saved the British Grid
Inside the control-room nightmare of 23 June: the day Britain's grid ran out of margin — and was saved by a phone call
For forty-eight minutes on the evening of Monday 23 June — a 33°C midsummer Monday — the frequency of Great Britain's electricity grid sat below 49.8 Hz, the boundary of normal operation. At 17 seconds past eight o'clock it touched 49.657 Hz, its lowest point of the year — though, as I wrote in the immediate aftermath, it stayed clear of the statutory floor at 49.5: this was never a blackout in waiting, and this article will not pretend otherwise.
Nothing switched off. No public warning was ever issued. And almost nobody noticed that, for 48 minutes, Britain's electricity system was running on a draining battery fleet, a gas margin of roughly eighty megawatts, and — finally — an unpriced phone call to Paris.
I've spent the last month reconstructing that evening minute by minute, from both sides of the Channel: Elexon settlement data down to individual generating units, NESO's one-second frequency record, the system operator's own trading books, and — via RTE's APIs — the French grid's answering records. Along the way I have also had the benefit of conversations with people who watched the evening from the other side of it — traders and market folk in several European markets, who saw the British grid's difficulties arrive on their own screens, and who cannot be named. Their fingerprints are on this piece in ways I won't itemise. That access shaped a deliberate choice: the mainstream telling of 23 June is relentlessly British — British data, British institutions, British questions. But half of this story happened in other countries' control rooms and other countries' order books, and I have tried to tell it from their side of the water too: as our neighbours and partners saw it, not merely as we did. What follows is the anatomy of a near-miss that the official record describes, accurately but incompletely, as a day when "the grid kept working."
It did. In the same sense that a driver who runs a red light and isn't hit keeps driving.
Act I — A grid already two-fifths gone
The story of 23 June begins weeks earlier, in maintenance calendars.
By the morning of the 23rd, 21.2 gigawatts of British generating capacity was on outage — 12.6 GW planned, 8.6 GW unplanned. Half of Dinorwig, Britain's biggest pumped-storage station, was away on nine-month overhauls. All four units at Cruachan sat at zero. Connah's Quay-2 had failed the previous day. And Sizewell B — the only nuclear station in the south of England — was entirely absent, both units on statutory outage since May: 1.2 GW missing from precisely the half of the country where the evening's shortage would land. And on top of the outages sat another 2.6 GW of standing partial derates — big CCGTs registered at one output and physically capable of less.
The cables to Europe were wounded too. IFA, the original French interconnector, had been running on one-and-a-half poles since a DC cable fault in March. Greenlink to Ireland was out entirely. And ElecLink — the interconnector through the Channel Tunnel — had a quieter handicap: planned 400 kV works in the Calais corridor had zeroed its export direction for the summer, leaving it a one-way street toward Britain. That morning the one-way street worked fine. By mid-afternoon it would become the day's strangest mystery.
One more thing the control room knew that morning, and it matters later: conditions on the other side of the Channel were tough too. France was fighting the same heatwave with a large slice of its nuclear fleet unavailable and river-cooled reactors carrying warm-water derates. Not a crisis over there — but not a neighbour with power to burn, either.
Against this backdrop, the day-ahead auctions did something that would define the evening: they scheduled Britain to export electricity to the continent, all evening, with the evening peak priced at £250/MWh. High — two to three times a normal June evening — but comfortable. The market saw strain. It did not see what was coming.
Act II — The sunny day that worked
Here is the first myth of 23 June: that too much solar power broke the grid.
The opposite is true. Midday — maximum sunshine — was the easiest part of the day. Solar delivered 15.2 GW, slightly more than forecast, and Britain had so much power that the system operator's problem was surplus: NESO spent the morning selling electricity at £76/MWh and managing low inertia, the classic headache of a renewables-rich grid at noon. The sun then set on schedule, exactly as predicted. Solar was the best-behaved actor on the entire system.
There was a second gift from the sun that morning, and it came through the cables. At midday Britain was importing around 5 gigawatts from the continent — much of it Europe's own solar surplus washing across the Channel at prices that made burning gas pointless. And here is the trap hidden inside that comfort: sunset is synchronised across Europe. As Britain's panels faded, so did France's, Belgium's and Holland's — and the same cables that had delivered 5 GW at lunchtime swung to exporting 3 GW by 19:00, an eight-gigawatt reversal in seven hours, right into Britain's evening ramp. France's own export cushion touched zero at 19:00 BST, the exact hour Britain's crisis began. Interconnectors are sold as diversification; against a continental sunset, they are correlation.
The morning's real story went unnoticed. At 10:00 and 10:40, two units at Pembroke — Britain's flagship gas station — failed forty minutes apart. Eight hundred and ten megawatts, gone. In a grid drowning in solar, the loss was absorbed invisibly and the market barely blinked. The lesson, which the evening would teach brutally: the size of a failure matters far less than its timing.
Act III — The repricing nobody heard
At 13:00, ElecLink stopped. Not commercially — the intraday book had it importing at its full 1,014 MW through the afternoon and into the evening. Not officially — its published import capability read No Restriction, 1,014 MW, in every declaration NESO carries. The metered flow simply fell off a cliff: 848 MW at half past twelve, 36 at one o'clock, then zero — and it stayed at zero for the next seven hours, through the entire build-up you are about to read. Nearly five gigawatt-hours of bought-and-scheduled French imports, physically absent, with no fault notice, no capability restriction, and no outage record on either side of the Channel to account for it. I call those the phantom seven hours, and you can only find them by holding the schedule book against the metering.
At 14:18 — seventy-eight minutes after a gigawatt of imports vanished from the wires — something remarkable happened in NESO's trading room: the operator bought its entire evening safety margin in a single burst — 111 forward trades, 5,700 MWh, executed within minutes of each other, at prices rising to £876/MWh. Three and a half times the day-ahead price. Hours before the event. Priced, it would turn out, almost exactly at what the evening settled at.
The control room, in other words, saw the nightmare coming at lunchtime — including, evidently, things the public data did not show — and quietly paid up for insurance. What it did not do was tell anyone. No Electricity Margin Notice was issued — not then, not ever that day. The market's last public forecast of the evening remained £250.
The gap between what the public market believed and what the system actually cost is the single most eloquent chart of the day. The same evening megawatt-hour carried three different prices depending on when you asked: £250 in the day-ahead auction, £561 in the intraday market as traders caught up through the afternoon, £800 at final settlement — and £1,032 for the marginal action at the peak. Each price was a snapshot of how much of the truth had leaked out.
And from mid-afternoon, every layer of the system thinned at once:
The heat shaved 2,247 MW off the operating gas fleet — 12.7% — and, counterintuitively, it cut deepest on the newest, most efficient machines. Keadby 2, Britain's most efficient CCGT, lost 10.6%. Pembroke's surviving units lost 13–16%. The 1990s veterans barely moved: their age had already been priced into their ratings years ago.
Wind ran 1.7 GW below a forecast that was — astonishingly — still rising. NESO's own updates pushed the evening wind forecast up to 7.1 GW during the day; the turbines delivered about 5. The most accurate forecast NESO ever had was the one published two days early.
Across the solar shires — Cornwall to East Anglia — 32 substations flipped, one by one between 15:00 and 18:00, from feeding the grid to drawing on it, handing nearly 7 GW of demand back to the transmission system as rooftop solar died with the light.
And at each hour boundary, the auction-scheduled export ramps hit the grid like hammer blows: at 18:00, four interconnectors swung 1,109 MW of exports on in five minutes, and frequency dropped an eighth of a hertz on the spot.
Sheffield Solar's PV_Live data lets you see the third of those bullets properly, and it deserves to be seen, because it demolishes the lazy reading of the evening. Between four and nine o'clock, the demand on NESO's screens climbed six gigawatts — which sounds like a nation coming home and switching on. It was the opposite. Add the solar back and underlying consumption actually fell five gigawatts across those hours — Britain was winding down for the evening. What climbed was not demand but visibility: eleven gigawatts of rooftop generation went down with the sun, and five hours of hidden consumption walked back onto the transmission grid to be served the old-fashioned way. The grid's hardest ramp of the year happened while the country was switching things off.
By early evening, the deliverable gas headroom south of Britain's congested transmission boundaries had collapsed from 8 GW at lunchtime toward a few hundred megawatts — a 40-to-1 collapse in seven hours.
The wind story deserves its own picture, because it is genuinely strange. Forecasts are supposed to converge on the truth as the horizon shortens. On 23 June they did the opposite: the two-day-ahead forecast was the most accurate NESO ever had, and every subsequent revision moved away from what the turbines would actually do — the on-the-day updates climbing to 7.1 GW, even 8 GW, while the fleet delivered five. Whatever the models saw building in the evening weather, the wind never showed up to match it.
Was the miss extraordinary? No — and that's the frightening part. Ranked against every day of 2026 so far, 23 June's wind forecast error was the 80th largest out of 198. A one-and-a-half-to-two-gigawatt evening miss happens roughly two days in five. Britain routinely runs forecast errors larger than its entire formal reserve standard — the SQSS requires securing 1,320 MW — and normally gets away with it because incidental headroom absorbs the difference. On 23 June, the incidental headroom was gone. The error was ordinary. The margin was not.
And wind was only half the forecasting story, because demand was misbehaving in the same direction. The day-ahead demand forecast ran about 800 MW under the actual evening, and at the worst half-hour of the ramp the miss reached 1,446 MW — another Cottam-and-a-half of shortfall, invisible until teatime; only the 16:48 revision, published as the squeeze was already on, finally caught the shape of the evening.
But the deeper problem with the demand forecast is not its error. It is its geography — it has none. The forecast NESO publishes, trades against, and calibrates is national: one number for an island whose crisis that evening was entirely about where the demand sat relative to full wires. A national figure can be nearly right and operationally useless at the same time — and on the 23rd it was: nearly right nationally, while the south east starved behind constraints the forecast cannot see. Nor can an outsider check the geography after the fact: I tried to reconcile NESO's day-ahead zonal demand forecast against the settled per-substation metering, and the two products don't share a basis a member of the public can bridge — the zonal forecast counts transmission demand by boundary, the settlement data nets off every rooftop panel behind each substation, and several gigawatts of embedded generation sit unexplained between them. Add the calibration culture — decades of tuning demand models against cold winter evenings, not hot summer ramps — and you have a forecasting system pointing its sharpest instruments at the wrong season and the wrong map.
The best an outsider can do is compare not levels but shares of the evening ramp — how the forecast distributed the coming surge across the country, against where the settled metering says it actually landed. Even that limited comparison is damning, and it can be summarised in two words: top-heavy. A word on the line I'm cutting along: B9 is NESO's Midlands to South of England boundary — the belt of wires crossing the lower Midlands roughly from the Severn estuary to the Wash, whose eastern end runs through the Spalding–Walpole–Bicker Fen circuits. Everything below it — South Wales, the South West, the Home Counties, London, East Anglia — is "the south" in what follows; it is also, not coincidentally, the line the evening's constraints bound against. NESO's day-ahead zonal forecast put 46% of the evening ramp north of B9 — 7% in Scotland alone. The metering's answer: Scotland contributed nothing — zero — and the country south of B9, the solar belt, took 79% of the entire ramp against a forecast of 54%. And top-heavy is not a neutral kind of wrong. Demand forecast in the north is the easy kind — it sits beside Peterhead, the wind fleet and the Norwegian cable, and it flatters your margin. Demand that actually lands in the south must be served through wires that were already full. A top-heavy forecast is, automatically, an optimistic one: the operator's map of the evening was tilted toward exactly the half of the island where serving demand is cheapest — and the evening arrived in the other half.
Walk the same comparison down the length of the island — cumulating the ramp boundary by boundary from the top of Scotland — and the two curves tell the whole story in one glance: the forecast's amber line climbs steadily southward, the way a winter evening builds; the metered line hugs zero the whole way down, dips negative through the North West and Yorkshire, and then leaps to 100% in the final step. The forecast expected Britain to switch on from the top down. Britain switched on at the bottom, and only at the bottom.
A caution on that northern dip, because the metering can't distinguish two very different stories and the substation detail says both were happening. Some of the "emptying" is embedded generation, not behaviour. Blyth in Northumberland pushed ~350 MW onto the grid all evening — that is Lynemouth, the 420 MW biomass station behind the substation, running hard. Grimsby West on the Humber swung 138 MW across the evening and flipped to net export by nine — and most of that swing is a single machine: the 80 MW KXP Immingham battery behind the substation, which was charging at four o'clock and discharging at 77 MW through the peak. Which is a pleasing thing to find in the demand data: the battery relay of Act IV, visible from the other side of the transformer, dressed up as a town switching itself off. But strip the embedded sites out and the behavioural story survives, and nowhere more cleanly than in the capital: the two biggest emptiers in the country were City Road and St John's Wood — central London's own substations, −207 and −108 MW, with no large embedded plant behind them. That is the Square Mile going dark on a hot Tuesday evening, measured at the transformer.
One cut deeper is available, because in one place the geographies line up almost perfectly: London, where the forecast's B14 boundary and the capital's settlement metering agree within 100 MW at four o'clock — same city, same basis. From there the day-ahead had London adding 1,138 MW through the evening: the classic teatime pickup of the winter textbooks. What London actually did was fall 472 megawatts. On a 33°C evening the capital emptied — offices dark, aircon spinning down, a million people in parks and beer gardens rather than kitchens — while the model, calibrated on cold January evenings, confidently switched on a city that wasn't there. A sixteen-hundred-megawatt round trip between expectation and reality, in the one place on the map where we can measure it cleanly.
Act IV — Forty-eight minutes
Frequency first crossed below the 49.8 Hz operational limit at 18:52:55. What followed was not one event but three distinct battles.
Episode one lasted eight minutes and was beaten back by a wave of balancing-mechanism dispatches.
Episode two began at 19:04 and ground on for 26 minutes — the longest — while Cottam Development Centre, a 400 MW gas unit on its first day back from ten weeks of maintenance, collapsed in flight: 318 MW, then 178, then 13, then nothing. Episode two ended at 19:30:00 precisely, in a single stroke: an Emergency Assistance instruction to the Netherlands' TenneT cut BritNed's exports from 750 MW to zero. One phone call, 575 MW. A lever that had been available for the entire 26 minutes.
Give the Dutch their due here, because the French rescue has the poetry and TenneT's part is usually forgotten: the first Emergency Assistance of the night was Dutch, twenty-five minutes before Paris, and it was the single cleanest action anyone took all evening — one instruction, executed to the second, worth a sixth of a hertz on the spot. TenneT's records show the cut being staged from 19:00, while episode two was still deepening.
And be clear about what the favour actually was, because it is easy to misread. Britain was exporting to the Netherlands that evening — the day-ahead auctions had sold the Dutch 750 MW of British power, and it was flowing. The Emergency Assistance instruction didn't send the Dutch anything; it let Britain keep power the Dutch had already bought, in the middle of their own evening peak. Seven hundred and fifty megawatts is roughly four per cent of Dutch demand, surrendered to the second.
So where did the Netherlands find it? Here I owe the reader honesty about the limits of the data. The Dutch fleet didn't move — Eemshaven, Hemweg, Diemen, ENECOGEN all held almost perfectly flat through the cut. The cut itself was staged over half an hour, notice enough for TenneT to cover the gap the way system operators cover known shortfalls: reserves, then intraday purchases, dissolving into a continental pool that is one synchronous machine from Lisbon to Poland. And inside that machine, attribution stops working. On a meshed AC grid, physical border flows never map cleanly onto anyone's contracts — every trail I followed toward a named rescuer dissolved, on inspection, into schedules that were already moving for their own reasons. The Dutch found the power the way the continent always finds power: everywhere, and therefore nowhere in particular. When Britain begged, the Dutch gave up their contracted dinner, and the continental plate absorbed the difference without leaving a receipt.
One mechanism, though, is legible — and it is the closest the data comes to showing where the Dutch cover came from. All afternoon, the Netherlands had been physically delivering 450–600 MW more to Belgium than the schedules called for. For precisely the hour containing Britain's frequency minimum, that over-delivery stopped: the Belgian flow snapped back to its schedule, and the power that had been slipping south stayed on the Dutch grid instead — roughly 450 of the 750 megawatts the Dutch had just surrendered from Britain. Then, once the worst had passed, the southward drift resumed and deepened, reaching 1,400 MW by late evening. Belgium itself, meanwhile, surrendered nothing: its own British cable, Nemo, kept drawing about 730 MW out of Britain all evening, straight through the frequency minimum.
Which brings us to the evening's most uncomfortable ledger entry. NESO did try to turn the other cables commercially: it spent about £1.6 million countertrading Nemo and Viking, and in the final commercial schedules both cables duly flip from export to import at 19:00. The metered cables never turned. Nemo kept exporting ~730 MW and Viking ~1,140 MW — commercial reversal, physical persistence; Britain bought its energy back on paper while the electrons kept flowing out. Add Ireland's cables and the arithmetic lands like a punch: at 20:00:17, as the frequency touched 49.657 Hz, Britain was physically exporting roughly 2.4 gigawatts to Belgium, Denmark and Ireland. Only the Dutch cut — the one favour delivered as physics rather than paper — and, minutes later, the French reversals actually turned flows around. The taxonomy of the evening is stark: the Dutch gave, the French acted, Germany underwrote — and a further 2.4 GW of contracted exports simply kept walking out the door, £1.6m of buy-backs notwithstanding. Norway's cable to the Netherlands, NorNed, sat at a published zero all day — no direct Norwegian help for the Dutch — though some of the power crossing the German border may well have been Norwegian water arriving the long way round, via NordLink; and Norway's hydro was in any case flowing to Britain directly, down North Sea Link, as the biggest single infeed on the British grid. Follow the electrons and the conclusion is unavoidable: this was not a bilateral favour but a continental one. German and Danish plant, Norwegian water, Belgian transit and Dutch forbearance, all rebalancing around one island's empty evening — most of it invisible in any British record.
It could not last, though. At 21:30 the Netherlands took its power back — BritNed's exports resumed at 725 MW, and Britain's frequency sagged 0.15 Hz on the spot. The evening's first IOU, cashed before midnight.
While the frequency trace fought its episodes, the control room was buying — and the buying tells its own story. Between three in the afternoon and midnight NESO accepted roughly £8 million of offers in the balancing mechanism, and the pattern is two markets stacked on top of each other. Along the top of the chart below, at £970–992/MWh, sits a row of open-cycle peakers — Grain, Taylor's Lane — the grid's most expensive conventional machines, run because they exist. Through the middle, the big CCGTs: Seabank alone collected £2.49 million across the evening at an average of £743. And underneath, a cloud of blue: hundreds of battery acceptances at £350–600 — Tesla's Clayhill at £466, SSE's Ferrybridge at £404 — consistently 30–40% cheaper than the gas being bought in the same half-hours. Cheaper still were the battery offers not taken: in the tightest hours, about 41% of in-merit battery offers were skipped, at an average price around £352, while gas cleared at £656 to £1,032. Some of that is prudence — a battery megawatt-hour used at seven can't be used at eight. Some of it is a dispatch system that still finds it easier to run a 1970s peaker than to trust ten thousand cells.
Behind all of it stood Britain's traditional keeper of bad evenings — Dinorwig, the pumped-storage cavern in Snowdonia — fighting at half strength: three of its six units were away or declared unavailable, and the three that remained ran flat through the crisis. The mountain did its job. There was just half a mountain.
Episode three was the deepest, and it was a race.
On one side: the battery fleet. Britain's batteries were magnificent that evening and it's important to say so — a staggered relay of discharges (a second shift of fresh cells arriving at 19:00 sharp, exactly as the first faded), 3.3 GW at peak, two-thirds of all balancing acceptances in the crunch, delivered with near-perfect precision, at prices — around £513/MWh in the crunch — some 30–40% cheaper than the gas bought alongside them.
The fleet's whole day is worth a chart of its own, because no other machine on the grid can draw this shape: discharging into the morning peak, swallowing 1.4 GW of the midday solar surplus, then swinging 4.7 GW within a couple of hours to meet the evening — a fleet of ten thousand containers behaving like one perfectly obedient power station.
But batteries hold energy, not just power, and the lower panel is the plot twist: the tank. The fleet had been discharging since teatime on the market's schedule, and by the frequency minimum the measured response of the dynamic-containment units — the automatic shield — had halved at the same frequency error. Over the whole evening the balancing-mechanism fleet discharged 6.3 GWh — 108% of a full charge. They did not underperform. They ran out — arithmetically, completely.
