The UK Energy Problem
An interactive model of Britain’s electricity system measuring how far current plans fall short and modelling how to close the gap.
Why we built this
Electricity demand is set to almost double by 2040, but current plans don’t provide enough power for cold, still winter days and the plants needed to address that take years to build, so Britain risks running short from the early 2030s. We built the model to test options for solving that problem and to understand how big and how urgent it is.
- Demand is set to surge. In NESO's central Holistic Transition pathway, demand nearly doubles, from about 291 TWh in 2025 to 564 TWh in 2040. In that pathway, which is built to meet today's net-zero targets including the vehicle and heating mandates, electric vehicles rise from 9 to 111 TWh and heat pumps and heat networks from about 1 to 44 TWh.
- AI computing comes on top of that. By mid-2025 about 140 data centres, seeking roughly 50 GW, were queuing for grid connections. The government's own forecast that Britain will need at least 6 GW of AI-capable capacity by 2030 implies roughly 40–50 TWh a year, already above NESO's central 37 TWh for all data centres in 2035. We build the model on a demand case of about 97 TWh by 2035 (why we chose that figure).
- Wind and solar need firm back-up. On cold, still winter days, renewable output can stay low for long periods. The grid then depends on dispatchable plant and imports. Britain has imported about ten percent of its electricity in recent years and NESO issued six capacity margin warnings this summer.
- Lead times are long. Large gas turbines come mainly from three manufacturers, whose order books are full four to five years ahead. Turbines ordered today would not arrive until about 2031 and a new combined-cycle station takes five to seven years from order to power. New nuclear takes longer and planning and grid connection rules cause further delays. Plants needed in the mid-2030s have to be ordered TODAY.
- Missing it has a cost. Without enough firm power, Britain risks losing AI and industrial investment to countries that can connect it, slower growth, greater dependence on imports and a less stable grid.
The model's own results underline the urgency. At the rate Britain is actually building, and even with the storage and interconnectors already in the pipeline, current plans miss the 3-hour reliability standard from the early 2030s under every NESO demand pathway (see below). Plants needed then take five to seven years or more to deliver, so they have to be ordered now.
What the model covers
The model is built on over 1,000 documented assumptions drawn from 100 cited sources and a record of nearly 3,000 existing power stations. For every year from 2030 to 2050, it replays each hour of the year across seven historical weather years and two severe cold-spell scenarios, with and without electricity trade with Europe, giving a total of 567 simulated years for each plan. A further 1,764 simulated years, with random power station breakdowns, test each plan against the official 3-hour reliability standard. Together these produce around 600 calculated results per year, from generation and costs to fuel use, emissions and supply shortfalls and over 12 million stored results.
Version 2 adds an estimate of the cost of new grid connections and wider network upgrades, including projects already under construction, shown alongside generation, storage and fuel. Over 2030–50 in the central case it adds about £361bn to current policy delivered, against about £173bn for the nuclear buildout, because offshore wind needs far more new grid than plant built on existing power station sites.
In the model, no plan that relies on wind, solar and unabated gas reaches 95% clean while keeping the lights on. Even building wind, solar and batteries at the Clean Power 2030 pace to 2050 reaches only about 90%. Holding gas to 5% leaves demand unmet, whilst letting it run to keep the lights on breaks the 95% commitment. Carbon capture on the new gas stations would close most of the difference, at an approximate additional cost of £100bn, but very little is planned and its cost is too uncertain to model. Only the nuclear buildout reaches 95% clean, by 2050, while meeting the reliability standard.
Version 2 also includes the interconnectors already approved (rising from 10.3 GW today to about 14.5 GW by 2033), Ofgem’s 16 provisionally selected long-duration storage projects (7.6 GW, 2029–2033) and the current limit on moving power from Scotland to England. Together they cover the gap until the mid-2030s. After that the shortages come when Europe is also short, and in cold, still spells lasting several days, longer than the new storage can run. They narrow the gap but do not close it.
1. If nothing changes
The explorer opens on current policy as Britain is actually delivering it. Over 2021–25 Britain added on average about 1.3 GW of offshore wind, 0.5 GW of onshore wind and 1.6 GW of solar a year. Counting the projects now under construction, the recent pace is about 2.2 GW of offshore wind, 0.55 GW of onshore wind and 2 GW of solar a year: roughly a quarter to two-fifths of the pace Clean Power 2030 needs. This plan carries that pace forward, with Sizewell C and the planned Wylfa reactors, the long-duration storage and interconnectors already in the pipeline, and a gas fleet kept near today’s 35 GW by replacing older stations like-for-like. Nothing else is added.
How fast Britain is building
| GW added a year | 2021 | 2022 | 2023 | 2024 | 2025 | Average 2021–25 | Under construction now | Pace used in current policy | Clean Power 2030 pace | Shortfall |
|---|---|---|---|---|---|---|---|---|---|---|
| Offshore wind | 0.87 | 2.59 | 0.89 | 1.41 | 0.51 | 1.25 | 11.85 GW | 2.2 | 6.5 | 3.0× |
| Onshore wind | 0.42 | 0.35 | 0.69 | 0.67 | 0.23 | 0.47 | 2.04 GW | 0.55 | 2.3 | 4.2× |
| Solar | 0.48 | 0.56 | 2.76 | 1.86 | 2.49 | 1.63 | 3.2 GW | 1.9 | 4.8 | 2.5× |
| Batteries | about 1.5 a year (6.8 GW operating at end-2025) | 1.5 | 9.9 GW | 2 | 3.6 | 1.8× | ||||
Britain would not have enough electricity, and would miss its carbon target.
Not enough power: hours a year when supply falls short
Extra firm power needed to meet the 3-hour reliability standard
Carbon target missed: clean share of electricity
- Not enough power: in the central case Britain is about 8 GW short of firm power in 2035, 25 GW in 2040 and 44 GW in 2050 (33–59 GW across the demand cases). Expected shortfalls reach 1,321 hours a year by 2050, against a standard of 3.
- Not clean: with gas filling the gaps, the clean share stalls at about 75% in 2040 and falls to 71% by 2050 as demand grows, far from 95%.
2. If the build problem is solved but policy stays the same
Current policy delivered is the best case for the current approach. Wind, solar and batteries are built at the Clean Power 2030 pace every year to 2050, gas is replaced like-for-like, and gas engines and open-cycle turbines are added until the 3-hour standard is met in every demand case. No new nuclear, carbon capture or hydrogen is assumed, because almost none is being built.
Build rate: gigawatts a year, today versus what current policy needs
That is three times today’s offshore wind rate, nearly five times onshore and two and a half times solar. It keeps the lights on, meeting the 3-hour standard every year in every demand case. But it still misses the carbon target.
Clean share: the target is still missed
Wind and solar cannot cover long, cold, still spells and four-hour batteries cannot bridge them, so extra turbines mostly add output when there is already a surplus: about 263 TWh is switched off in 2050. The clean share stops at about 90%. Total cost over 2030–50, including grid connections and network upgrades, is about £1,666bn.
3. A different plan: the nuclear buildout
The nuclear buildout keeps the renewables already built or contracted but adds no more. Instead it builds a fleet of small modular reactors, from 2039, and one large reactor in 2037, rising to about 50 GW of nuclear by 2050, with gas engines, open-cycle turbines and four coal stations as backup while the reactors arrive. It is sized to the same 3-hour standard and meets it every year in every demand case. It reaches about 96% clean by 2050 in the central case and costs about £1,195bn over 2030–50 including grid, £471bn less than delivering current policy, because reactors on existing sites need far less new grid and run whatever the weather.
The three plans compared
Clean share of electricity, three plans
Total cost 2030–50, including grid
| Central demand, 2030–50 | Current policy | Current policy delivered | Nuclear buildout |
|---|---|---|---|
| Meets the 3-hour standard? | No, from 2031 | Yes, every year | Yes, every year |
| Expected shortfall, 2050 (standard 3 hours) | 1,321 h | Under 3 h | Under 3 h |
| Clean share, 2050 | 71% | 90% (about 96% with carbon capture) | 96% |
| Generation, storage and fuel cost | £938bn | £1,305bn | £1,021bn |
| Grid connections and network upgrades | £141bn | £361bn | £173bn |
| Total | £1,079bn | £1,666bn (about £1,766bn with carbon capture) | £1,195bn |
| CO₂ 2030–50 | 1,036 Mt | 642 Mt | 820 Mt |
Over the whole period the nuclear buildout emits more CO₂ than current policy delivered (820 against 642 Mt), because its reactors arrive through the 2030s and 2040s and it relies on gas and some coal until they do; by 2050 it is the cleaner system. The range across demand cases is in the explorer.
Solving the problem
We are so far behind in energy generation, that this problem cannot be solved by supply side changes alone. We are simply too far behind to be pursuing EV targets for vehicles and converting to heat pumps whilst supporting an industrial strategy and widespread AI adoption. Priority to AI buildout and economic growth has to come above other demands for power generation. These models assume there is at least some reform to improve the current planning timelines.
How much AI demand are we planning for?
The AI expansion demand case plans for about 97 TWh a year of data-centre demand by 2035: NESO’s central 37 TWh plus 60 TWh more, or about 11 GW running day and night. It is a planning choice, not a forecast, and it is above every NESO estimate. The gap chart above uses NESO’s own pathways, so the finding that current plans fall short does not depend on it.
The evidence
- The government’s own forecast. The UK Compute Roadmap (July 2025) forecasts that Britain “will need at least 6GW of AI-capable data centre capacity by 2030”, three times today’s data-centre capacity. Run at 75–95% of capacity, 6 GW is roughly 40–50 TWh a year, around double NESO’s central 2030 figure for all data centres (23 TWh), and it is for 2030, not 2035. The roadmap does not say whether 6 GW is computing load or whole-site capacity including cooling, so the TWh figure is approximate.
- Independent evidence to Parliament. LSE’s Centre for Economic Transition Expertise used a stylised 10 GW at 90% use, which is 79 TWh a year, with a 20 GW high case. Our 97 TWh sits between the two.
- The connection queue. By mid-2025 about 140 data centres were seeking roughly 50 GW of grid connections. NESO judges that close to half of pipeline capacity may be highly speculative or duplicative, but even a modest share would be very large.
- US forecasters keep revising up. BloombergNEF expects US data-centre demand to reach 106 GW by 2035, up from about 25 GW operating in 2024; that forecast is 36% higher than its own estimate eight months earlier. NERC, the North American grid reliability body, now expects peak demand to rise by over 224 GW in ten years, 69% more than it projected a year before, with data centres accounting for most of the increase. US national load forecasts have now been revised up for three years running.
- NESO’s latest view is lower. Its September 2026 Ten Year Outlook cut data-centre demand to about 15 TWh in 2031 and 24 TWh in 2036 (about 22 TWh in 2035), citing updated utilisation data and project attrition. That is well below its own 2025 scenarios and far below the government’s 6 GW forecast. The two official views now differ by a factor of two to three, which is why we plan for the higher end rather than the lower.
| Estimate | Data-centre demand | Year |
|---|---|---|
| NESO Ten Year Outlook 2026 (latest, revised down) | about 22 TWh | 2035 |
| NESO 2025 ten-year forecast (before the cut) | 33 TWh | 2035 |
| NESO 2025 scenarios, lower (Falling Behind) | 20 TWh | 2035 |
| NESO 2025 scenarios, central (Holistic Transition) | 37 TWh | 2035 |
| NESO 2025 scenarios, higher (Electric Engagement) | 41 TWh | 2035 |
| Government 6 GW forecast, at 75–95% use | 40–50 TWh | 2030 |
| LSE stylised case, 10 GW at 90% | 79 TWh | — |
| Our AI expansion case | 97 TWh | 2035 |
| LSE high case, 20 GW at 90% | about 158 TWh | — |
Why we picked 97 TWh
- The government’s forecast already exceeds every NESO figure, and it is a 2030 minimum. It implies tripling capacity in five years. If growth continues after 2030, demand in 2035 will be higher still.
- Forecasters have repeatedly underestimated this demand. In the US, the main forecasts have been revised up year after year as data-centre orders arrived faster than expected. NESO has moved the other way, cutting its figure for 2035 from 33 TWh to about 22 TWh. We would rather plan for the upward trend seen in the largest market than assume Britain will be the exception.
- It sits inside the independent range: just above LSE’s 10 GW case and well below its 20 GW high case.
- The reforms are designed to attract computing investment, through faster connections and routes for large loads to bring their own power. The supply plan has to be sized for that policy succeeding.
- The risks are not symmetric. Firm power takes five to seven years or more to deliver. Planning too low means investment goes elsewhere and the grid runs short; planning too high means some capacity is built early. Large loads that bring their own firm power reduce the public requirement one for one, so if demand arrives with its own power, the public plan shrinks.
Using the explorer: what is instant and what needs a run
Instant
- Fully calculated plans: current policy at today’s build rate, current policy delivered and the nuclear buildout, under four demand pathways: NESO’s three and AI expansion. The explorer opens on current policy.
- For each, year by year to 2050: capacity, generation, imports, fuel use, emissions, costs, supply checks across seven weather years and two cold-spell stresses, and hours against the 3-hour reliability standard.
- If you edit a plan's build orders, its fleet, construction timetable and construction spending update immediately.
Needs a model run
- Hour-by-hour results for an edited plan: generation, shortfalls, reliability hours, fuel, emissions and running costs.
- These show as "Hourly recalculation required", so out-of-date figures are never displayed.
- You can save an edited plan from the explorer as a file and open it again later. I may add the python script later so users can build their own plans and save.
Explore the model
Start from current policy and its gap, then compare delivering current policy in full with the nuclear buildout across four demand pathways: capacity, generation, costs and security of supply, including the 3-hour reliability standard.
Open the explorerMethods and evidence
The audit report sets out every method, source and assumption. The workbook lists the inputs and results.
Audit report Workbook