Reversing scorched earth policies in the electricity sector Part 2

In this article I will address two crucial questions that must be considered when planning for a future electricity system outside the confines of Net Zero policies. How much generation capacity will we need in 2030 or 2035? How will changes in electricity demand affect the trade-offs between (possibly) cheap but intermittent sources of generation and dispatchable but perhaps inflexible types of generation?

The issue of future demand has come to prominence because of reports that developers have expressed interest in obtaining grid connections for data centres with a total capacity of 50 GW to meet expected demand for AI applications. Many of these proposals won’t get beyond the stage of being gleams in the eyes of over-optimistic developers. Still, even if only one-half were to go ahead that would represent an increase of more than 50% in peak demand for the GB electricity system. Further, data centres put huge pressure on electricity systems to maintain very high levels of reliability. They cannot be served by solar or wind farms, unless an enormous investment is made in storage or backup generation.

Making good forecasts of future electricity demand is notoriously difficult. In terms of system planning, it is necessary to allow for 10, even 15, years of project planning and construction plus at least 10 years of operation. An experienced colleague of mine once commented about the CEGB in the 1970s and 1980s that they were equally bad at forecasting future demand and building new power stations. Fortunately, the errors cancelled out, so the system finished up with adequate reserves of capacity. Still, we cannot rely on offsetting errors to save us in future.

Is it likely that there will be a boom in building data centres like that reported for the US? The plans for new US data centres amount to about 150 GW for a population of about 345 million or 435 MW per million population. That demand covers training new AI models as well as serving many users outside the US. Neither element is a major factor in the UK. Stripping out such factors, the maximum plausible demand in the UK will be well under 200 MW per million population or around 14 GW. Still a lot, but not 3 times that figure.

Next, we should take account of basic economics. Plans to build new data centres in the US are concentrated in locations where power prices are low, and very definitely not in California or New England. The US average for industrial electricity prices in 2025 was $83 per MWh. In comparison, the average for the UK was about £240 per MWh or $322 per MWh, nearly 4 times the US average. Would developers contemplate building large numbers of data centres when they must pay such high electricity rates?

The idea is ridiculous. The requests for grid capacity would only become serious projects if the level of electricity prices for large users were to fall by two-thirds or three-quarters. The subsidies required to offer large quantities of electricity with a very high level of reliability at such prices would be astronomical. Under the current system, the burden would fall on other electricity users, which would lead to huge political and economic resistance.

Equally, why build new power plants when transporting data traffic is much cheaper? The unit cost of installing new submarine fibre optic cables is £30,000 to £40,000 per km. Such cables can carry 200,000 Gbps (gigabits per second). One such cable could cope with the plausible demand from 20 million households.

The North Sea Link (NSL) – the power interconnector from Northumberland to Norway – has a length of 720 km. A single fibre optic cable would cost less than £40 million for that distance whereas the NSL cost over £2 billion at 2025 prices and has a capacity of 1.4 GW. You can build 50 fibre optic cables with the capacity to handle more than 10 times the plausible level of total UK traffic for data centres in 2035 for the same capital cost as a single NSL interconnector. In simple cost terms it is far cheaper to transport traffic to data centres in Norway – or other countries with low power prices such as Sweden and France – relative to the options of importing power from Norway or building generating capacity in the UK.

The primary constraint on transporting data is not cost but the time that is required for data to travel to and from data centres – what is called latency by network engineers. The amount of latency that is acceptable for internet traffic varies hugely with type of use. High speed financial traders want a latency of 10-15 ms, while keen gamers get grumpy if their latency is above 30 ms. On the other hand, a latency of 100 ms would be acceptable when dealing with a chatbot or an AI program that is drafting a letter or memo.

In fibre optic cables, bits travel at about two-thirds of the speed of light. The time taken for a return journey over a fibre optic cable with a length of 1,000 km is about 10 ms (milli-seconds). Each time a packet is handled by a router or other networking equipment adds from 5 to 10 ms to the latency. High speed traders want direct routes over short distances with a minimum number of hops. Some potential applications of AI – such as medical diagnostics – may be sensitive to latency delays, but most applications will not be significantly affected by sending traffic to data centres over 500 km to 1000 km. What will be more important will be potential delay due to limits on processing capacity at the data centre. That is why large users of AI are so concerned to increase the amount of data centre capacity available for their applications. Building lots of large data centres in the UK makes no economic sense.

Stepping back from the special pleading from different interest groups, total final demand for electricity in the UK has declined by an average of 1.1% per year since its peak in 2005. Thus, a naïve projection forward to 2030 yields a forecast of total final demand of 259 TWh, a decline of 5.4% relative to 2025. A more sophisticated forecast that takes account of the large increase in electricity prices since 2020 on demand by sector yields a slightly lower figure in the range 250-255 TWh for 2030.

In its Clean Power Action Plan NESO adopts a significantly higher figure for total demand in 2030 of 287 TWh with the increase coming from a switch to electric vehicles and heat pumps. Their impact is partially offset by an assumed reduction of 20 TWh in residential electricity use due to improvements in energy efficiency. Nearly 18 months after NESO’s plan was published most of it seems even less plausible than it did at the outset.

The crucial conclusion is that my argument in Part 1 of this series that total final demand for electricity in 2030 can be met without any contribution from intermittent solar and wind generation is even stronger. That result was based on meeting actual demand in 2025 which was 5-10% higher than reasonable forecasts looking five years ahead. With lower overall demand, the level of peak demand is likely to be lower, and the reserve margin of spare capacity will be larger.

This leads on to the second question raised at the beginning of this article. How much will it cost to dispense with the contribution from solar and wind generation?

The financial incentives for solar and wind generation are entirely different from those that apply to dispatchable generators. In general, solar and wind plants incur very low or zero operating costs per MWh of output. Most generators receive either guaranteed prices or top-up payments that mean they receive positive revenue even when market prices are negative. So, wholesale market prices do not affect whether they operate or not. All that matters is the amount of sun or wind required for generation.

Removing solar and wind generation from the system will push up the demand for dispatchable generation. Inevitably, this will increase wholesale market prices, while reducing the payments made to solar and wind generators. Using a standard statistical model, I estimate that on average in 2025 the wholesale market price increased by £3.95 per MWh for each 1 GWh of solar and wind generation removed from the market.[1]

Based on this model, I have estimated the total value of generation in 2025 at (i) average market prices for each hour [Actual], (ii) adjusted market prices if all intermittent generation had been removed from the market [Alt_1], and (iii) adjusted market prices paid only for additional supplies to replace solar and wind generation [Alt_2]. The key difference between scenarios Alt_1 and Alt_2 is that under Alt_1 existing dispatchable generators receive a huge bonus through the increase in the wholesale price required to bring additional dispatchable generation to replace solar and wind generation. This is pure profit to them and is strictly unnecessary to ensure that total demand is met. However, avoiding the payment of such a large bonus would require a significant change in how the electricity market works. This will be discussed in a later article.

The total value of electricity supply at market prices under the three scenarios are as follows:

Actual scenario: £22.3 billion

Alt_1 scenario: £35.1 billion

Alt_2 scenario: £28.1 billion [£5.8 billion for new dispatchable generation]

These figures show that the total value of generation at market prices in 2025 would have increased by £12.8 billion if all dispatchable generation were paid the same price after removing solar and wind generation from the market. This increase is still lower than the total cost of subsidies for intermittent generation which amounted to £25.2 billion in 2024 at 2025 prices.[2]

Equally important, the cost of replacing intermittent generation could be reduced to £5.8 billion by avoiding the payment of a large – and unnecessary - bonus to existing dispatchable generators. In fact, the estimate of £5.8 billion is a maximum figure because, again, there is no reason to pay the full increase in the wholesale market price for all replacement dispatchable generation. If the single buyer model was adopted in a sensible manner, the cost of replacing solar and wind generation would be close to £3 billion for the 2025 level of final demand.

Getting rid of ETS (emission trading system) permit requirement would more than fully offset the increase in the wholesale market price required to replace solar and wind generation. In 2024, the cost of ETS permits increased the average wholesale price by about 29%. The calculations above assume that ETS system remains and becomes more expensive because additional dispatchable generation will increase the demand for ETS permits. If the ETS system was abolished, the gross cost of replacing intermittent generation would be only £2.8 billion, i.e. a total value in the Alt-1 scenario of £25.1 billion.

Crucially, and finally, consider the combination of three reforms: abolishing the ETS system, replacing all intermittent generation with dispatchable generation, and adopting a single buyer system of pricing for dispatchable generation. This would reduce the total cost of generation at market prices by £2.3 billion or a little more than 10%. In addition, almost all the costs of support for renewables plus additional balancing costs would be eliminated a potential saving of about £11.8 billion per year.

In summary, without Net Zero policies the total final demand for electricity is likely to fall by 5-10% by 2030. Meeting this demand without use of solar and wind generation is both feasible, and this option should get less expensive over time. Relying on current market arrangements to replace solar and wind generation will be expensive and quite unnecessary. By eliminating the ETS permit system it would be possible to reduce overall generation costs even after replacing solar and wind generation. That in turn will allow for a large reduction in support for solar and wind generation that will lower the overall costs of power generation including support for renewables by 35-40%.

[1] The details are as follows. Market data for 30-min settlement periods were combined to give hourly averages of the market price, intermittent generation, total system load, and other indicators. Such data is highly autocorrelated. To allow for this, the model was estimated using first differences, i.e. the change in the market price relative to the previous period was assumed to be a function of (a) the change in the total output from solar and wind generators, and (b) the change in the total system load. Some autocorrelation remains even after taking first differences, so the model was estimated using what is known as the Prais-Winsten method. The data are very noisy but the estimated change in the market price per GWh of intermittent generation of £3.95 is very well determined with a standard error of £0.18.

[2] An estimate of renewable subsidies for 2025 cannot be compiled until after the end of the fiscal year 2025-26.

27/02/2026 - Typo corrected concerning the cost of a fibre optic cable crossing the North Sea.