How to accelerate the UK’s AI revolution

The arrival of AI has seismic implications for energy systems; the question of how to power demand from data centres is set to command policy and industry attention across the globe in the next few years. Tech companies are looking for quick ways to procure large amounts of (ideally clean) power, in the context of slow grid expansion in Western economies - creating significant tension between their desire to build and the inability for current energy systems to enable hyperscalers to do so.

While the US has seen the most notable surge in such demand, leading to diverse and sometimes carbon-intensive energy procurement strategies, the UK is facing similar challenges. It currently hosts 523 data centres, the third highest of any nation globally[1], with at least nine more in development.

The Government is acutely aware of the need to balance the growth of its AI sector with its ambitious clean power targets - and the establishment of the AI Energy Council and 'AI Growth Zones' (AIGZs - designated hotspots for future data centres) this year signifies an intent to foster mutual growth without one sector impeding the other.

The UK currently stands at a crossroads, as it decides how to power its data centres of the future - and at what speed. Levers it can pull include:

Siting assets in locations with existing grid capacity - and finding ways to accelerate their grid connections - is the idea behind the UK government’s ‘AI Growth Zones’, some of which have recently been announced, with further decisions still pending. In an already highly constrained grid, there’s a physical cap to the number of these zones that can exist. Whilst an important and welcome short term step, AIGZs are unlikely to meet the predicted increases in demand from AI over the coming years alone.

Meanwhile, expanding the grid takes a long time and is expensive. The ‘Great Grid Upgrade’ is underway - the largest overhaul of the grid in generations, involving the building of new infrastructure and upgrading existing systems. The UK needs to build five times more transmission infrastructure in the next five years than was built in the last three decades[2] - inevitably leading to uncertainty about our ability to hit this target. The UK no longer has a great track record when it comes to building things quickly. Gone is its manufacturing base - and construction productivity rate.

This makes circumventing the grid and connecting directly to new sources of power, via microgrids and private wires, increasingly appealing - especially as fixed costs are loaded onto consumer electricity bills. Such investment decisions are occurring against a backdrop of profound shifts in UK power markets, where centralised planning from the system operators is replacing wholesale market signals. Grid defection allows investors to avoid legacy system costs and expand capacity in step with local demand, in ways that the national system cannot[3].

Options open to hyperscalers include procuring nuclear power and leveraging the UK’s first small modular reactors (SMRs), using renewable sources of energy, and seeking out quick gas connections - the latter of which is gaining increasing attention from data centre operators[4]. These decisions are all taking place in the context of the Clean Power 2030 target and the need to grow the UK’s AI capacity and unlock economic growth in the short term.

Key questions examined in this analysis:

What we did

The modelling is based on supporting a 120MW data centre - which sits at the top-end of GB data centre deployments. Access the modelling to play around with the assumptions here.

Key Findings

The renewables + gas scenario 2 typical yearly cost is 43.4% less than the nuclear SMR scenario 1, when considering capital and operating costs, annuitised over the operational lifetime of generation assets. If you remove solar generation (scenario 3), in a land-constrained scenario, this cost saving falls slightly to 42.1% - although emissions from greater gas use increase slightly. The difference in cost is driven by the relative CapEx and OpEx costs of the different scenarios. Note we haven't included any premium that would be associated with procuring power from an SMR already in development, nor do we account for the impact of carbon prices on gas operational costs over time.

A 95% renewable microgrid with 5% gas backup - in line with the UK’s Clean Power 2030 target - was modelled at almost a third (31.7%) lower cost than scenario 1 in today’s prices. In this model, the gas is restricted to just under 80MW (2/3rds the size of the data centre) and the model correspondingly chooses a larger battery for storage, and increases the size of wind and solar technologies.

Figures 3-6 demonstrate how the proportion of each generation source varies in scenarios 2 and 3 by day of the week, and month of the year. Generally, we see offshore wind meeting the majority of load requirements, with solar, batteries and gas supporting. Overall, gas is used on average 20.4% across the entire year in scenario 2, and 22% in scenario 3 where there is no solar generation.

These outputs highlight gas outcompeting batteries today as the cheapest supportive source of power in the microgrid set-up; our model optimises for the lowest-cost outcome. Here, there are three things to consider:

What happens when there’s a ‘dunkelflaute’ (a period of low wind), a lack of sunshine, or a combination of the two?

The UK is well placed for clean power: windy days are typically more cloudy, whereas sunny days tend to be less windy. Just 2% of days in a typical year have both low wind generation and low solar generation[7].

However, given the importance of reliability of power for data centre operators, in figures 7 and 8 we have demonstrated how during periods of low renewable generation which typically occur during shoulder seasons (here, August - September), a combination of gas and batteries account for the majority of the data centre load in its place, sometimes for multiple days.

What do hyperscalers care about? Beyond continuous power, which is an implicit assumption in our modelling of each scenario, key areas are cost, speed, reliability and carbon impact - as implied in figure 9.

1. Cost

The modelling demonstrates the cost-competitive nature of renewables + BESS + gas versus the estimated cost for the UK’s first SMR

In June 2025, Rolls-Royce SMR was selected by the UK Government as the preferred bidder to develop the UK’s first three SMRs[8]. £2.5 billion of funding was pledged - the sum Rolls-Royce estimated a first 470MW SMR would cost in 2021-22[9]. We used this figure for our CapEx estimations. Since that figure was published, interest rates and supply chain costs have risen - and costs are very site-specific and only become clear as planning and consenting processes progress.

The government has said that it will reach a project Final Investment Decision (FID) by the end of 2029. This will include funding to support site access and site-specific design. This means that some of the more time-consuming aspects of building new nuclear projects, namely site-specific regulation and relevant licensing, would not begin for any successful SMR design until after 2029[10].

In a 2022 submission, Rolls‑Royce stated its power will cost “around £50/60 per megawatt hour” in 2012 prices - which it re-confirmed was its position in early 2025[11]. Whilst noting industry scepticism around power prices below $180/MWh from a first-of-a-kind SMR[12] (~£133/MWh), we used £55/MWh in our modelling (again, with reference to 2012 prices). Even under what many consider optimistic cost assumptions for SMRs, our modelling shows the renewable scenario is substantially lower in terms of combined CapEx & OpEx costs.

Whilst nuclear will no doubt play an important role in the UK’s future energy mix, our modelling suggests it is an expensive option as a primary solution to meet new energy demand in the coming decade, even assuming new SMR developments can successfully deliver power for these estimated prices. To date nuclear has demonstrated little evidence of cost learning, with global construction costs remaining high and in some cases increasing over time. In the UK, nuclear costs have historically trended upwards. New modelling has recently found the true cost of the Sizewell C power station in Suffolk could be tens of billions of pounds higher than official government estimates, once financing costs are factored in - reaching a total possible sum of £100bn[13]. The government’s most recent estimates for the project were £38bn in real 2024 prices to build - a material increase from the original projections of £20bn in 2020[14].

SMRs are intended to reduce financing costs and mitigate many of the risks associated with onsite construction through their modular design, though this remains an expectation rather than a proven outcome. There are three SMRs that are operational globally: one in China, two in Russia, and a fourth currently under construction in Argentina. Costs are overrunning for all four projects from 300% to 700%[15] - as demonstrated in figure 11.

Meanwhile, renewables are proven technologies available for deployment today, and they’re cheap. Costs have plummeted, while nuclear costs have remained stubborn (or increased): between 2010-2022, global costs plummeted 90% for batteries[16], 89% for solar PV and 69% for onshore wind[17] - in the UK, costs for offshore wind and solar have fallen by approximately 70%[18] and 76%[19] respectively in the last decade.

2. Speed

Renewable microgrids could be deployed in around half the time of nuclear SMRs (~5 years vs 10 years) and with much greater certainty.

We estimate that microgrids relying predominantly on offshore wind could be built in approximately five years. The key advantage is that generation assets do not require NESO or offshore transmission operator (OFTO) engagement on grid-level consenting and modelling periods, which can take several years. They would be required to meet licensing standards under the Electricity Act, which may involve navigating non-standard regulations and regulatory reform.

We estimate approximate timelines for a large offshore private-wire wind farm as follows:

Part one:

Site selection to FID = ~36–42 months. This includes 6-18 months required for a private export cable consent, licence-exemption compliance and onshore planning rights. We’ve used existing offshore wind farms to approximate timings:

Part two:

Part two: FID to COD (Commercial Operation Date) = ~28-38 months. This is grounded in examples of the approximate amount of time it took for large offshore wind farms to be built in the UK:

We anticipate the construction of offshore wind farms as the most time-intensive part of the microgrid build-out - with other generation sources and cabling rolling out in tandem. For a 100MW private-wire solar farm in the UK, we anticipate a faster timeframe than offshore wind buildout: ~18–24 months to secure consent, following a ~9–12-month pre-application phase, followed by ~12–24 months for construction and commissioning. These estimates are anchored in recent UK examples: Cleve Hill Solar Park (373MW) had a 18-month consent timeline[26] and 27-month build out[27], and Shotwick (72 MW) had a 20-month consent timeline[28] and was constructed in a record 10 weeks[29]. Many BESS projects are completed in 6-12 months due to their modular design, but the development and planning phase can take a year or more[30].

These timelines rely on smooth consenting decisions and permissive regulatory structures. There are ongoing government efforts to compress consenting and development timelines - and the deliberate fast-tracking of strategically important industrial applications could reduce timelines further.

There are far greater unknowns when it comes to SMRs. Rolls-Royce’s first SMR is due to come online in the mid-2030s[31] - taking roughly a decade to build and operationalise. Figure 9 shows the timelines involved in each of the world’s existing SMRs, comparing projected and actuals, and highlighting the serious challenge of meeting that target on time. Not one has come close to its construction schedule[32]. This serves as a cautionary flag about this technology’s ability to play an efficient role in powering UK data centres in the near term.

3. Reliability

Data centre operators are looking for uninterruptible power supplies; reliability and high availability is central to business cases. Our model outputs a microgrid that meets consistent yearly data centre demand without needing the grid. To minimise risk to data centre operators, a low/no-utilisation grid connection can be built that is only used in emergency scenarios.

If a grid connection could be secured on a pay-per-use (or penalty-per-overuse) basis then this model could ensure reliability for the data centre operator and guarantees for the network operator. For example a 120MW data centre receives a 120MW-capable grid connection - but there is an explicit agreement that because the microgrid can cover the majority of demand there will be a penalty for overuse of e.g 10MW. Securing consents for this residual grid connection would add development time, estimated to be around six months.

4. Carbon impact

Nuclear power also requires back-up - and has a historic reliance on fossil fuels.

Scenarios 2 and 3 explicitly model gas as a backup source, to achieve the most cost optimal model of powering a data centre and to ensure high reliability and availability of data centre uptime. Note: the carbon intensity of open-cycle gas turbines (OCGTs) in the UK is 460 tonnes/MWh[33]. In practice, all data centre operations would rely on some level of backup to ensure this availability. In Scenario 1, we did not model a backup source. This is due to limited data on how often this would be used. However, it is unlikely that nuclear-powered data centres would be 100% carbon-neutral in reality.

Big tech companies such as Amazon, Microsoft and Google have started investing in nuclear power as a strategic lever to meet their 24/7 carbon-free energy commitments. Unlike variable renewables operating in isolation, nuclear offers consistent baseload power, aligning with the reported 24/7/365 operational demands of large-scale data centres and AI workloads - whilst noting emerging observations about low data centre utilisation rates[34].

However, there’s limited evidence that nuclear-powered data centres would be carbon-neutral - by virtue of the Energy Availability Factor (EAF) of nuclear power. Despite the common assumption that it is always-’on’ and available, unplanned and planned power outages means this isn’t the case. The reliability of nuclear and its potential role in Europe’s decarbonisation strategy was called into question in 2022, when France’s nuclear contributions were interrupted due to extended maintenance shutdowns and curtailment due to weather issues - resulting in 40% availability of maximum capacity for about a month[35].

With operational data on SMRs still absent, the best data available on the consistency of output from existing nuclear power plants is the International Atomic Energy Agency. They publish the EAF of all nuclear reactors in commercial operation in the last three years by country. The UK’s nuclear EAF averages at ~75% - highlighting a 25% power gap. At present, this gap is typically filled by a combination of gas-fired generation, interconnectors and renewables, depending on availability. In a scenario in which nuclear is meeting constant load requirements from a data centre, fast-ramping supporting power sources will be key - and many data centres are already leveraging gas whilst waiting to secure a grid connection[36]. Over thirty enquiries have been logged by UK gas networks from data centre developers in recent months, with reports that operators are “willing to pay the extra” to build gas plants to avoid grid delays that cost millions[37].

Finally, the need for back-up power must be considered. Regulations mandate nuclear power plants restart within 10 seconds of a grid failure, leading to the common adoption of diesel generators as a source of back-up power by nuclear plant operators[38]. Whilst SMRs will theoretically rely less on back-up sources compared to traditional power plants because of their design (lower thermal load, passive cooling systems etc), most early-deployment SMRs are expected to retain diesel generators as a primary or secondary emergency source to meet safety requirements. The Rolls‑Royce SMR design explicitly includes consideration of diesel generators as a back‑up option for loss of external power: ‘Rolls-Royce SMR Limited is considering back-up systems for providing power, including the use of and options for diesel generators. Diesel generators are a well understood and commonly used technology for this purpose. Other technologies and combustion fuels are also being considered.’[39] Whilst the UK’s environmental permitting regime places a cap on the amount of back-up diesel generator data centre operators can use in the UK, we speculate that all the scenarios modelled involve a certain level of carbon emissions.

1. Whole system costs

A key advantage of a pay-per-use (or penalty-per-overuse) grid connection model is that because it limits its effective size, it also curbs the system costs that are added to household bills - whilst providing hyperscalers with further assurances that data centres can run reliably.

An additional advantage of having the generation grid-connected even over a smaller grid connection means that these assets can provide flexibility to the grid by participating in markets and services, which can decrease system costs and therefore bills.

In future, there may be the potential to grow this grid connection over time as the grid expands and more capacity becomes available, depending on location. This would require smart regulation that protects customers during the grid integration process - but could allow data centres and generation sources to participate in more grid markets and services, with benefits to consumers.

Data centres are potentially valuable sources of demand-side flexibility. In order to avoid the opportunity cost of that lost flexible demand, if it is possible to successfully integrate data centres back into the system, in a way that makes optimal use of offshore wind and supports flexibility, the benefits should make their way back to consumers and their bills. This flexibility would involve adapting operations and scheduling AI load to periods when the grid is greener, enabling data centres to become a valuable balancing tool in a grid powered by renewables. Whilst this technique is already used by some tech players, innovators are exploring the potential for this to work for data centres at scale[40].

2. Location constraints

We have not spatially resolved our modelling. This means that we haven’t accounted for geographic and land use constraints that might limit where these future data centres and supporting renewable generation could realistically be sited - though this is a potential area for future work. While SMRs will be more flexible than conventional reactors, they will also be subject to spatial constraints. Despite having a smaller footprint than wind and solar, they will likely need to be sited in locations that satisfy regulatory, safety and cooling requirements, and where suitable grid connections exist. Their modularity may increase flexibility compared to conventional reactors, but siting options remain limited in practice. The comparative spatial demands of the alternative options would be a fruitful area for future work.

This modelling is not designed to provide a blueprint for sustainable UK data centres, nor for the optimal siting of data centres in the future. Inspired by analysis undertaken in the US[41], we aim to start a conversation about the potential of renewables to help rapidly power the AI race on this side of the pond.

The UK’s data centre expansion represents both an urgent growth opportunity and a defining test of energy system design. In the short term, the Government’s imperative is clear: accelerate AI growth and data centre development to unlock near-term economic gains. A potential £44 billion of Gross Value Added to the UK economy from data centres is on the table between 2025-2035, if capacity can be increased above its recent annual trend growth rate[42]. This requires infrastructure that can be deployed rapidly, reliably and affordably.

SMRs are an emerging technology that may be quicker to deploy than nuclear has been historically, but it is unproven. SMR’s greatest proponents are not yet able to point to a single at-scale working example in the world[43]. The previously mentioned examples in Russia and China, while technically SMRs, are not comparable in terms of power output or economics. If pace is paramount, our starting position should be to deploy proven technologies.

Renewable-powered data centres can mitigate the cost, time and risks that hyperscalers seek to avoid. They could provide a solution that allows building of data centres at pace - and there are increasing signs that recognition of this reality is taking hold globally. China has recently mandated that new data centres in its national hub nodes must use at least 80% green electricity[44], excluding nuclear, demonstrating its strategic preference for renewables and storage to meet 24/7 demand at scale.

Realising the potential of renewables in the context of data centres requires targeted regulatory reforms, integrated planning, and an approach to microgrid development that ensures long-term fairness in system costs. But the window of opportunity is narrow. Decisions taken by hyperscalers, investors and policy-makers in the immediate years ahead will shape the UK’s energy and digital landscapes for decades to come.