How do we keep the lights on with 12 GW net firm capacity at risk of retirement by 2030? Speech to the Institution of Power Engineers, 13 November 2025

Good afternoon everyone. I’d like to say I’m delighted to speak today about the CCGT retirement risks we face, but that would be a lie, because while it’s always nice to be with Institution of Power Engineers, it’s far from a delightful topic. However, it’s one that’s increasingly keeping me awake at night, because my analysis to date suggests we face real risks of rationing later this decade and into the early 2030s.

The shape of today’s power grid

In the past 25 years we’ve been racing to build more renewables. We now have 32 GW of wind and almost 20 GW of solar. This is backed up by 32 GW of gas generation, and another 12 GW of nuclear and biomass, just under 5 GW of hydro and pumped storage, and just under 7 GW of batteries.

Current interconnector capacity is 8.8 GW, but imports can’t be relied on if continental systems face the same weather pattern, causing corresponding stress events….with the exception of France and Norway our connected markets both share similar weather to us and are following a similar wind-led energy strategy.

France relies on an aging nuclear fleet and has a more temperature sensitive grid than us – during cold spells, we often export to France rather than import from it. And Norway is going off the idea of exporting electricity altogether since it has seen increases in power prices and power price volatility. Some in the Norwegian parliament want to cancel the interconnectors it has with Britain and Germany.

So we have roughly 64 GW of dependable firm capacity against 48 GW of peak winter demand.

Of course, no form of generation is 100% available so even if a technology is firm and dispatchable, it will still have maintenance outages and the occasional unplanned trip. This is accounted for through de-rating factors, and after applying these adjustments the current amount of available firm capacity is 50.5 GW, which gives a spare margin of just 2.5 GW based on firm power.

In reality there is generally some contribution from wind, but this can be very low – on several occasions this year, the actual output of the installed 32 GW was below 1 GW and at times as low as 200 MW (which, amusingly is Grok’s estimate of the amount we could generate if everyone in the UK blew very hard into all the wind turbines!)

Despite higher amounts of wind on the system, in 2025 wind output is down, not just in Britain, but across Europe. Solar of course contributes nothing to the winter peaks which are always at night. After the sun sets. A depressing number of people still need this explaining to them!

NESO, Ofgem, DESNZ and just about anyone with an opinion on this says we need to hang on to all of that 32 GW of gas to keep the lights on, on those cold still winter days when demand is higher and renewable output low to non-existent.

So it’s pretty unfortunate that a third of the fleet was built in the 1990s and is getting a bit long in the tooth. My analysis shows that it’s not in fact 10 GW of gas capacity that’s nearing retirement, but 12 GW.

We may gain a couple of GW of smaller units, particularly open cycle turbines, and 3.2 GW of new nuclear at Hinkley Point is expected to open in the early 2030s, but by then almost 5 GW of AGRs will have closed. The net retirement risk is therefore about 12 GW of firm, dispatchable capacity in the late 2020s and early 2030s.

Let me repeat that – firm power equivalent to over a quarter of current winter peak demand could be off the grid in the next few years. And there is no concrete plan to replace it.

Not only are we sleep walking into a capacity crisis, we’re trying to make it worse. The Government wants to decarbonise everything, and as it’s now widely accepted that hydrogen will play a minimal role, and many, including the Climate Change Committee are sceptical about the role carbon capture will play.

This means electrification is seen as the primary route to decarbonising heating, transport and whatever industry we have left. Indeed de-industrialisation, in power grid terms, is something of a blessing in disguise, taking some pressure off tight margins.

The Government also wants to AI data centres to be built in Britain, having designated them as critical national infrastructure – that could add another 6 GW of demand by 2030 according to recent Government analysis.

So in this presentation, I will go through the scale of the CCGT-retirement challenge; why the winter risk and demand control issues escalate; how this plays out in different demand-scenarios (status quo, electrification only, AI only, electrification + AI) to 2030; and finally to offer commentary on why technological solutions such as hydrogen or CCS are unlikely to scale in time, and why grid and equipment lead-times are a serious obstacle.

Condition of the CCGT fleet

Let’s examine the numbers. Britain’s gas-fired CCGT fleet includes many units commissioned in the 1990s, now approaching 30 years of operation, and reaching the end of their design life. This means increasing maintenance costs, decreasing reliability margins while facing market, regulatory and environmental pressures. In my analysis, I have identified about 12 GW of net at-risk capacity by 2030 if we factor in new small OCGTs and nuclear.

A further 6.3 GW is at risk from 2030 – 2035.

This is not just about quantity but quality. As those units age we can expect to see increased forced-outage rates, higher maintenance downtime, and perhaps de-rating due to wear, less flexibility to ramp down/up, and less ability to provide ancillary services.

So the retirement risk is not simply in lost MWs, but in the reduction of system resilience.

Anyone who has read my recent work on voltage control should find that concerning. Scotland is facing particular challenges as once Torness closes there will be only one synchronous generator – Peterhead – left to stabilise voltage in the country.  

The Spanish grid operator recently requested emergency powers to stabilise voltage in the south of the country over similar worries – a very uneven distribution of conventional generation has left the grid in the south of the country so weak it is concerned about further blackouts.

Peterhead commissioned in 2000 and does not appear to have had any meaningful engineering upgrades since. In the past year it has had an unusually high number of unplanned outages, according to REMIT transparency data.

It logged 30 distinct unplanned incidents in the year to 12 November 2025, of which ~4 appear to be full trips and ~26 were partial derates. The incidents cluster in November-24, May-25, July-25 and October-25. Using last-published timestamps as a conservative proxy for duration, the cumulative unplanned downtime could be as much as 2,800 hours.

More worrying is that Langage, which was built a full decade later, has seen even higher levels of unavailability. In the past 12 months there were 58 distinct unplanned incidents, of which around 17 appear to have been full trips and 41 partial derates.

Although individual events were generally short – averaging about 12 hours – the sheer frequency meant that Langage experienced roughly 700 hours of unplanned downtime in total.

Industry sources suggest turbine-vibration issues which is consistent with the REMIT data. This underlines a wider point: even the more modern plant is being stressed by cycling and grid conditions, so simply relying on newer CCGTs to carry winter security is optimistic.

Winter risk  – why is this retirement problem particularly  acute in the winter?

In winter, demand peaks due to shorter days, more lighting needs, highly variable wind speeds impacting wind generation, little to no solar, and higher heating loads. As demand rises while dispatchable capacity falls, the margin shrinks. Under a cold spell, even modest variation in demand or supply can trigger stress.

Winter stresses the provision of generation as well as increasing demand, reducing both capacity and availability. These risks are amplified when much of the gas fleet consists of aging plant.

• For example there’s a risk of brittle fatigue from thermal cycling. Modern F-class turbines were designed for relatively steady baseload or 1-2 daily start-stop cycles. Today’s two-shifting with its multiple starts per week means thermal stress on turbine blades, rotors, and heat recovery steam generator tubes. In cold weather the metal temperature delta is larger at start-up, producing higher thermal gradients and more risk of cracking or deformation
• Auxiliaries including instrument air, fuel-gas heaters, cooling-water lines, condensate drains, are all points of failure in freezing conditions
• Condenser cooling water and closed-cycle cooling systems can suffer ice formation at low ambient temperatures, especially for air-cooled condensers
• As temperatures drop, lubricating oil becomes more viscous, making it difficult for it to circulate quickly to all critical components which creates accelerated wear on vital parts like bearings and gears, and increased load and strain on oil pumps and starter motors increasing the risk of trips on start-up
• Automated start-up sequence controls can be less reliable during cold weather operations. Minor faults or unexpected data from frozen instrumentation can upset the complex control logic, causing protection systems to engage
• Where cold, dry weather creates low humidity, static charges can build up on non-conductive materials which can discharge to sensitive electronic components, causing operational disruptions or permanent damage

Then there’s fuel supply risk: CCGTs require a specific, consistent gas inlet pressure to the combustion chamber for efficient and stable operation. Lower inlet pressure directly limits the mass flow rate of gas that can be supplied to the gas turbine’s combustion chamber leading to a notable decrease in efficiency. In extreme cases of low gas pressure, CCGTs may have to significantly reduce generation or even shut down to protect equipment.

During cold spells, gas system pressure is lower because of heating demand – more gas is taken off the system than is added. The 2018 “Beast from the East” weather event saw significant impacts on the GB gas grid’s inlet pressure, causing reductions in power output and efficiency in the CCGT fleet.

Even as renewable capacity grows, the experience of wind-lulls in the UK (multi-day periods of very low wind) is well documented. Studies have found that such events are frequent and can last days.

If 12 GW of firm capacity is retired, the head-room – that is, the buffer between available capacity and demand – disappears and a shortfall of up to 10 GW emerges on low wind days. This would inevitably lead to demand control, but raises the risks that unexpected outages of the remaining plant or an interconnector import shortfall would require sudden increases in demand reductions, and in such a scenario, under-frequency or voltage stability risks increase.

Older CCGTs may see more forced outages and maintenance trips. If several ageing units trip simultaneously in more extreme cold winter conditions, the absence of spare capacity means that increased demand control (which is a polite way of saying pre-planned load-shedding) becomes more likely. In short: the larger the “hole” in dispatchable capacity, the greater the probability of larger, trip-induced supply shortfalls leading to further demand intervention and possible blackouts.

Beyond capacity, CCGTs provide inertia, frequency response, voltage-support and other services. As they retire, new replacements (batteries, renewables) provide fewer of those services or do so in different ways. Inverters must use current for all the services they provide – if they use current to support voltage they cannot use it to power loads. Large numbers of inverters, programmed the same way in line with the Grid Code, could respond to the same voltage event by reducing active power creating a drop in frequency and a risk of blackouts.

The CCGT retirement risk is not just a capacity shortfall but also creates risks of system instability, frequency excursions, voltage collapse and so on, all of which carry the threat of blackouts. We’re entering a period in the coming winters where demand stress, supply variability, and capacity retirements converge. The system will be much more exposed than many people realise.

Capacity shortfalls increase under different demand scenarios

We can now consider how this exposure plays out under three demand scenarios to 2030, and a fourth combining two: (i) status quo demand growth; (ii) electrification only; (iii) AI-driven demand only; (iv) electrification + AI.

Under a status-quo demand scenario, we don’t really see any demand growth. In recent years, despite increasing population, de-industrialisation has pushed electricity consumption lower:

• Demand peaks may remain broadly at current levels at around 50 GW (last year it was 47.4 GW, but Amira is currently forecasting 47.5 GW next week)
• Retiring 12 GW of firm capacity means that if that loss is not replaced by equivalent firm capacity, margins disappear into a shortfall on low wind days
   – while the system may still function, under regional demand control, any large forced outage or interconnector import shortfall (eg due to cold and still weather on the Continental) could trigger increased demand-control actions or elevated blackout risk
• Demand-control would become routine in low wind conditions
• The system operator would need to procure more reserve/ancillary services to bridge the gap, for example paid for demand-side response, increasing cost for consumers

Until a few years ago “triad avoidance” used to deliver around 2 GW of peak shaving, however Ofgem deemed triads were no longer an appropriate way to allocate network costs as it was unfair to households. This 2 GW of industrial and commercial demand response was lost. At the same time, the Demand Flexibility Service was introduced, and it was hoped I&C consumers would migrate to it. They didn’t so we went from 2 GW I&C demand response under triad avoidance to 200 MW domestic response under DFS. Getting back to GWs of peak shaving would need much higher prices which would feed through to bills.

Now imagine a scenario where heat-pump and electric car deployment accelerates and, industrial electrification progresses, causing demand to increase – a 7-10 GW increase would be a credible scenario. This would mean:

• Peak winter demand would rise to perhaps around 60 GW by 2030, widening the cold still weather supply gap
• The probability of multiple risk events (eg cold snap, wind lull, interconnector import drop, CCGT trip) increases materially
• Demand-control would be inevitable on low wind days and would likely affect more consumers

Next we consider a third scenario where electrification stalls but we see increased demand from AI data centres.

Here we assume that data‐centre growth, AI computing loads, and high‐performance computing create an additional demand load for electricity of some 6 GW. Here the impact is essentially the same as in the electrification scenario but the low wind shortfall is 1-4 GW smaller

Finally, if we consider a scenario where both electrification and deployment of AI data centres boost demand, clearly we end up in a far more stressed condition.

• Peak demand could plausibly rise by around 15 GW to around 65 GW
• Higher numbers of people would be impacted by still weather demand control with even higher system costs as NESO has to buy even more services to try to support the grid

However, my view is we’re not going to see material increases in electrification to 2030 and whatever we do see will be offset by de-industrialisation. While I do think there will be AI data centres, I think they’re going to rely on behind the meter generation, so impact on the grid and the capacity margins will be minimal. So the status quo demand scenario is the most likely in my opinion.

Why replacing retiring plant is difficult

Of course, the assumption would be that retirements would be offset by the construction of new generation. Historically it has been possible to build a CCGT from scratch in 18 months after breaking ground. But this is impossible in today’s market. Not only have permitting timescales widened, grid connections can take a decade, and there are 7-8 year lead times for delivery of new GTs.

Even a new rotor on an existing plant can take 5 years to deliver, and parts for more modest upgrades to hot gas path equipment can take 12-18 months to get to site.

For a new CCGT to come online to offset retirements to 2030, it would need to be not just sanctioned already, but have taken FID and ordered all the equipment. There are no projects that meet this description.

Even if generation can be built, connection to the transmission system can require reinforcement, network upgrades, possibly new substations, lines, and sometimes local connection issues. These network upgrades often face planning /consent delays and take years – it’s not uncommon for projects to be quoted connection dates in the mid-2030s. Without the transmission ready, a new CCGT cannot reliably participate in the system as firm capacity, unless it can use a pre-existing grid connection.

There is, of course, the narrative that hydrogen-fired gas turbines or CCGTs with CCS will provide the future firm capacity. But realistically by 2030 the scale of hydrogen or CCS in the UK generation system will remain very limited. Some generators have projects they claim to hope will open by 2030 but not having taken FID this is highly unlikely.

In the past year, NESO and the CCC have both adjusted their outlooks to show a minimal role for hydrogen – various hydrogen production and pipeline projects in Europe have been cancelled, and the British hydrogen village trials were terminated due to public opposition.

The CCC has said it sees minimal role for carbon capture.

New CCGT investment is subject to market risk  – if investors believe policy is moving away from gas because of net-zero trajectories, they may be reluctant to commit – indeed, Langage and South Humber have 15 year refurbishment contracts under the capacity market for 2028-29 onwards but appear not to have taken the final investment decision – delivery seems unlikely.

The retirement of old plants may accelerate, but the replacements may lag which will make everything a whole lot worse.

Likely reliance on demand interruptions and demand control

Given the emerging capacity gap and the elevated risk of rationing and blackouts, the implications for system operators, policymakers and consumers are significant. Demand interruption and control measures will become central to balancing the grid. These might include:

• Enhanced interruptible demand contracts for large industrial and commercial users
• Dynamic load-shedding schemes (triage of loads in extreme events)
• Greater use of demand-response aggregators offering capacity at short notice
• Price signals and variable tariffs that encourage reduction of peak demand
• Selective load curtailment in low wind, cold weather conditions

In effect the system operator may have to rely more on demand-side action to maintain balance rather than purely supply-side flexibility. That’s really bad news – utilities should serve consumers, not the other way round, and while consumers may be compensated for peak shaving, the costs of this will be added to bills. Economic output would also be adversely affected if industry is forced to reduce activity, potentially cutting total output.

Recommendations for policy-makers

For winter operation, planning needs to be more conservative. Reserve margins need to reflect the retirement of CCGTs and AGRs and potential extra demand. They also need to match cold weather demand with cold-weather generation and de-rate all technologies to cold weather rather than average performance. Growing renewables is all very well – although my views on their usefulness are pretty well known – but the system needs firm, dispatchable, controllable capacity (gas, hydro, nuclear) that can operate at full output when needed.

Replacing dispatchable generation with intermittent generation won’t cut it.

Policymakers will need to decide how much risk they are willing to accept and how much cost they are willing to incur to maintain reliability. Just this morning I was sent a study which showed that in addition to the 11 people who died as a result of direct factors in the Iberian blackout, there were around 160 excess deaths over the 2 days affected, so blackout risks are not to be trifled with.

Given the long lead-times for new firm capacity, policymakers must give clear, consistent signals to investors that firm-capacity is still valued. If the market believes gas is “out” and no firm-capacity revenues will flow, investment will fall short. The potential retirement of 12 GW of CCGT by 2030 requires a replacement strategy that must be started now, not in at some future date.

Given the risks we face, here are some practical recommendations:

1. Accelerate and simplify planning for new firm dispatchable capacity with the explicit goal of replacing the retiring 12 GW by 2030
2. Prioritise shorter lead-time options such as OCGTs and even coal if necessary to keep the lights on
3. Grid reinforcement and network connection programmes must be accelerated – the existing reform process is moving too slowly
4. Winter-stress scenario planning must assume lower head-room, higher demand and elevated forced-outage risk
5. Realistic demand-response programmes with much larger capacity must be developed, including interruptible contracts and dynamic tariffs to ensure that large-scale demand-response is ready to deploy
6. Avoid over-reliance on unproven technologies (hydrogen, CCS and long-duration storage)
7. Send long-term investment signals that firm capacity is still needed and will be remunerated
8. Explain to businesses and households that increased demands (heat pumps, EVs, AI data centres) require corresponding investment in reliability and may impose cost or behaviour change
9. While interconnectors and storage will help, they cannot be the only answer – storage can’t currently provide multi-day capacity and interconnectors depend on continental supply which may also be stressed in cold snaps
10. Condition-monitoring and maintenance of the ageing CCGT fleet must be maximised: any avoidable forced outage in the remaining units worsens the margin. Asset-owners may need incentives to invest in life-extension and this needs to be delivered as soon as possible

In conclusion, the retirement of approximately 12 GW of 1990s-built CCGTs in Great Britain by 2030 represents a major transition risk for our electricity system. Combined with variable renewables output, long lead-times for new firm assets, and the potential growth of new loads through electrification and AI, the margin for error is thin.

If we do nothing, we may find ourselves in a situation where the lights remain on, but only through more frequent demand-intervention, higher consumer cost, and increased system risk. If we act now, accelerate firm-capacity investment, prepare demand-side programmes, and maintain the flexibility and robustness of the system, we might be able to minimise the impact, but there is little recognition from NESO, DESNZ or Ofgem of the scale of the challenge.

That complacency increases the risks that the first time we actually see a shortage results in a blackout, because the control room may be caught unprepared. NESO has a poor track record of forecasting the size of wind output and the size and timing of peak demand. Getting this wrong on a marginal day, could put us at increased risk of blackouts.

We saw this on 8 January this year: NESO over-estimated wind output, under-estimated demand and incorrectly predicted the timing of the peak. It made a risky decision to replace Rye House (CCGT) with Dinorwig (pumped hydro) through the peak because the former was seeking very high prices in the Balancing Mechanism. To save money, Rye House was ramped down to zero just before the demand peak.

This was against a backdrop of various system margin warnings and a request made by NESO to Energinet to return a bipole of Viking back from maintenance early. Plant is particularly unreliable when returning from maintenance so at a time of system risk, NESO took the decision to run its only GW scale fast response asset instead of holding it in reserve, just so it could swap out an expensive CCGT for an hour or so.

This type of complacency significantly elevates blackout risks. Of course NESO must take account of costs to consumers, but it needs to do so sensibly. What it did on 8 January was not sensible.

.

The picture is bleak, and time is running out to materially reduce the risks. Without action the costs both in terms of economic output and consumer hardship will be significant.

I’m in discussions about a commission to write an in-depth report into these risks. If anyone has any insights they would like to share, perhaps about assets you’re familiar with, I’d welcome all input.

Thank you.

Q&A