Can the UK Grid Capacity Cope With AI Data Centres and Their Power Requirements?
Over the past year, UK policymakers and regulators have quietly started to admit something the data centre industry has known for a while: AI-scale compute is on a collision course with the power system and AI data centres UK grid capacity is becoming a critical constraint.
Ofgem now says that around 140 proposed data centre projects in Great Britain are seeking grid connections with a combined demand of about 50 GW of power capacity – more than the country’s current peak electricity demand of ~45 GW. If every one of those schemes were built and operated at full load (they won’t be), data centres alone could roughly double Britain’s peak electricity requirement.[1][2][3]
That “datacentres will double electricity use” line is therefore not just pub talk – it is rooted in Ofgem’s own connection queue data. The reality is more nuanced, but the direction of travel is clear.
AI Data Centres, UK Grid Capacity and Power Demand: Where Are We Starting From?
A few key baselines help frame the discussion:
- The UK’s current data centre footprint is still modest in system terms:
- Oxford Economics estimates that new UK data centres will add around 6.2 GW of IT power by 2030, more than doubling today’s ~2.9 GW installed base.[4]
- NESO’s Clean Power 2030 and subsequent scenarios now project UK data centre electricity demand rising to ~22–30 TWh by 2030 and up to ~70 TWh by 2050 in high-growth cases – up from 3.6 TWh in 2020.[6]
To put that in context, UK electricity generation in 2024 was about 228 TWh. So by 2030, data centres could easily be in the 8–10% of national electricity demand range, and a much larger share of commercial consumption.[6]
Internationally, the IEA and others put global data centres at ~1–1.5% of electricity use and ~0.5–1% of CO₂ emissions today, rising as AI workloads scale. The UK is on the sharper end of that trend because it is actively courting AI infrastructure – through “AI Growth Zones”, NSIP status and a forthcoming National Policy Statement for data centres.[7][8][9][10][11][12]
So the “doubling the grid” soundbite is best understood this way:
- Connection queue: ~50 GW of prospective data centre demand vs ~45 GW peak national load – a theoretical near doubling of peak demand if everything were built and ran flat out.[13][14][1]
- Realistic outcome: only a fraction will be built; but a 5‑10x growth in data centre electricity demand this decade is entirely plausible given NESO and government scenarios.[15][4][6]
Why AI Data Centres Are So Power Hungry
Traditional colocation facilities already draw 10–50x more power per square metre than a typical commercial building. AI and high‑performance computing push that further:[4]
- High rack densities: average rack ratings are already in the 7 kW+ range and trending up; AI-optimised racks can run at 30–80 kW.[16]
- Round‑the‑clock loads: hyperscale and AI centres tend to operate with very high utilisation factors, making them near baseload from the grid’s perspective.
- Cooling overheads: even with improving PUEs (hyperscalers now near ~1.1–1.3), cooling and auxiliaries still add 10–30% above “pure IT” load.[11][12]
The result is that individual sites can get very big, very quickly. Two proposed UK hyperscale/AI campuses – Elsham (Lincolnshire) and Cambois (Northumberland) – are each sized at around 1 GW of electrical demand, roughly the output of a nuclear unit. A single such campus, running on today’s grid mix, can emit in the order of 0.8–0.9 MtCO₂ per year.[16][17][18][19]
How Do You Power an Extra 20–50 GW of Digital Load?
1. The stressed grid: connection queues and locational politics
Ofgem’s consultation on demand connections makes it clear that data centre requests have “surpassed even the most ambitious demand projections”. Key issues:[13][2][3]
- Connection queues: demand-side projects (including data centres) have jumped from 41 GW to 125 GW of contracted offers between late 2024 and mid‑2025.[13]
- Network constraints: most of the UK transmission system was not designed for dozens of 100–1000 MW loads clustered around London and the South East. Hence the policy pivot towards:
In practice, that means grid strategy is now front‑and‑centre in planning; developers are expected to show credible phasing, substation land, and active engagement with DNOs/NESO.[7]
2. Bulk power: renewables, nuclear and gas
a) Grid renewables and PPAs
The default route is to connect to the grid and “green” consumption via:
- Long‑term power purchase agreements (PPAs) with wind and solar – e.g. Telehouse using a PPA linked to the 630 MW London Array offshore wind farm to cover a large share of its Docklands campus demand.[22]
- Corporate PPAs and RECs/GoOs to match consumption with renewable output, sometimes on a 24/7 time‑matched basis in the case of hyperscalers.[23][24][25][26]
PPAs are now a core tool: AWS, Google, Microsoft and Meta have each contracted multiple GW of renewables globally to cover data centre loads, and UK operators are following that pattern.[24][25][27][11]
Limitations: PPAs improve the carbon accounting and can add genuine new capacity if structured with “additionality”, but they don’t solve local grid capacity constraints. A London-area facility backed by a Scottish wind PPA still needs copper and transformers in London.
b) Nuclear (large and SMR)
The UK government is explicit about using advanced nuclear to support data centres:
- AI Growth Zones are being steered towards areas “with new clean power generation such as North Wales, the home of our first small modular reactors”.[8][13]
- An “Advanced Nuclear Framework” is pitched partly at powering factories and data centres.[13]
Nuclear is attractive from a data centre operator’s perspective: high-capacity-factor, low‑carbon baseload, and potentially dispatchable in the form of SMRs co‑sited with campuses. But timescales are long (mid‑2030s and beyond) and political/financing risk is non‑trivial.
c) Gas and peakers
Despite all the net‑zero rhetoric, many data centre projects still look to gas-fired generation for either primary or backup power:
- Gas turbines and engines are proven, flexible and can be dropped in behind the meter.
- From a carbon perspective they are significantly worse than grid electricity in the UK, where average grid carbon intensity was about 129 gCO₂e/kWh in 2025/25, versus >180–200 gCO₂e/kWh for unabated gas at point of use.[28][29]
Gas is therefore a short‑term “safety valve” but a long‑term liability in a tightening planning and ESG environment.
3. Backup and resilience: from diesel to batteries and hydrogen
No customer, regulator or cloud provider will tolerate “we lost your training run because of a local power cut”. So backup power is non‑negotiable – and it is here that a lot of the local air quality and carbon controversy sits.
Today’s norm: UPS + diesel
Most facilities still rely on:
- UPS systems – providing seconds to minutes of ride‑through and power quality conditioning.
- Diesel generator sets – sized for the full IT and cooling load, providing hours to days of backup depending on fuel storage.[30][31][32]
Technically, this works very well. Politically and environmentally, it is increasingly toxic:
- Diesel emits CO₂, NOₓ and particulates; routine testing can be a significant local air quality issue.
- Planning decisions are starting to turn on this – for example Edinburgh refused a data centre scheme in part because of concerns around conventional diesel backup and its compatibility with local climate goals.[33]
The transition: batteries + hybrid architectures
Battery energy storage systems (BESS) are now being integrated at much larger scale:
- BESS extends backup from minutes to tens of minutes or hours, enabling facilities to ride through short interruptions without firing diesels and to manage peaks.[31][34]
- Hybrid solutions using UPS + BESS + a reduced diesel fleet allow operators to:
Several UK and Irish sites already co‑locate significant BESS with data centres, often participating in grid services markets as well as providing resilience.[22][31]
Hydrogen and fuel cells
Hydrogen is emerging as the poster child for “clean backup”:
- Teledata in Manchester is installing a 1.2 MW hydrogen‑ready fuel cell microgrid to supply clean, resilient power alongside existing battery storage.[27]
- The proposed 600 MW Havering campus in East London includes a hydrogen fuel cell plant and HVO (hydrotreated vegetable oil) backup generators in place of diesel.[35]
- Companies such as PowerCell are launching dedicated hydrogen fuel cell systems targeted at zero‑emission backup and distributed power for data centres.[36]
- Network operators in the UK (e.g. Wales & West Utilities) explicitly see green hydrogen as a potential primary or backup power source for data centres in constrained regions.[28]
Fuel cells driven by green hydrogen offer:
- Zero local emissions (apart from water and low‑grade heat).
- High efficiency and modularity.
- The ability to run either as backup or as part of a continuous on‑site generation strategy.
But they shift the problem upstream: you still need a reliable, low‑carbon hydrogen supply chain, which is far from mature at UK scale. Using grey hydrogen simply relocates emissions to the SMR/ATR plant.
Greenhouse Gas Impacts: How Bad Could It Get?
At project scale, the numbers are uncomfortable:
- The proposed Elsham AI data centre is expected to consume about 3.7 TWh/year, with projected operational emissions of ~857,000 tCO₂/year on today’s UK grid mix.[17]
- A large Google data centre project in Thurrock has been estimated at >500,000 tCO₂/year of emissions, comparable to several hundred short‑haul flights per week.[18][19]
At system scale:
- UK data centre electricity use may rise from 5 TWh in 2023 to >25 TWh by 2030 in some scenarios, with further growth thereafter.[15][6][4]
- Using a 2025 grid intensity of ~129 gCO₂e/kWh as a rough benchmark, a 26 TWh data centre sector would emit around:[29]
26 TWh × 129 gCO₂/kWh ≈ 3.4 MtCO₂/year
That is not catastrophic in economy‑wide terms (the UK emitted ~400 MtCO₂e in 2022), but it is one of the few demand sectors expected to see emissions growth rather than decline this decade. If AI growth and data centre build‑out outpace grid decarbonisation, emissions could be substantially higher, as some global analyses suggest.[10][12][37][38]
Campaign groups are therefore pressing for:
- Data centres to be explicitly included in UK carbon budgets and net‑zero planning.[15]
- Developers to demonstrate that projects will not increase net UK emissions – for example by funding equivalent new renewable generation rather than relying on unbundled certificates.[26][18]
Can We Square AI Ambitions With Net Zero?
The emerging consensus in the research and policy literature is:
- Yes, but only if:
- Data centre energy efficiency (PUE, server utilisation, cooling) continues to improve.
- New loads are sited intelligently – close to abundant low‑carbon generation and water resources, not just fibre routes.
- Renewable and nuclear build‑out is accelerated in the same regions where AI data centres are growing.[9][12][39][26]
- On‑site and contracted clean power (PPAs, co‑located generation, fuel cells) genuinely add new low‑carbon capacity.
- Diesel is rapidly displaced by batteries, HVO and hydrogen for backup; planning regimes make that shift unavoidable.[40][31][33][36]
If, on the other hand, data centres chase the quickest grid connections and the cheapest short‑term backup (gas and diesel) while government treats their electricity demand as an unplanned add‑on, the sector risks:
- Locking in several MtCO₂/year of avoidable emissions in the UK.
- Forcing NESO to keep more fossil plant on the system for longer.
- Undermining the credibility of the UK’s commitments to a largely decarbonised power system by 2030 and net zero by 2050.[12][18][15]
Strategic Takeaways
For UK policymakers, investors and operators, a few implications stand out:
- Grid planning and digital strategy are now inseparable. AI Growth Zones, NSIP status and connection reforms are all attempts to retrofit a strategic lens onto what started as piecemeal private investment.
- Power is the new planning permission. The gating factor for UK data centres is less often land or fibre than firm, low‑carbon MW at the right node.
- Backup is becoming a licence‑to‑operate issue. Diesel is moving from “necessary evil” to “planning risk”, especially in urban and peri‑urban locations. Expect rapid movement towards hybrid BESS‑diesel, BESS‑HVO and ultimately hydrogen fuel cell solutions.
- GHG performance will shape winners and losers. Facilities that can demonstrate genuinely additional renewables/nuclear, low PUE, heat reuse and clean backup will be far easier to permit, finance and insure than those leaning on the residual gas and diesel system.
- Hydrogen is plausible, not panacea. Fuel cells and hydrogen backup can materially cut local and system‑wide emissions – if they use green hydrogen and are integrated into a broader regional energy strategy. Otherwise they risk becoming another greenwashed bolt‑on.
For a UK that wants both a globally competitive AI sector and a credible net‑zero pathway, the question is no longer whether we build the data centres, but where, how fast, and on what energy system. The answer will be written as much by NESO’s transmission build‑out, PPA structures, and hydrogen infrastructure decisions as by whatever the next generation of chips looks like.
For a deeper dive into how hydrogen could support AI data centres and UK grid capacity, see my analysis of UK hydrogen policy.
Sources
- [1] New datacentres risk doubling UK electricity use, regulator says – link
- [2] UK’s Data Centre Boom Could Break The Grid, And That’s A Big Problem – link
- [3] UK National Power Grid can’t keep pace with AI expansion, Ofgem warns – link
- [4] The UK’s data centre boom: growth trends, drivers, and the rising power challenge – link
- [5] Estimate of Data Centre Capacity: Great Britain 2024 – link
- [6] UK data centres and AI hosting will need a lot of energy – link
- [7] Planning for data centres in 2026 – What’s changed and what will matter most – link
- [8] UK Government Policy Paper – Delivering AI Growth Zones – link
- [9] Data centres: turning UK policy into reality means securing power and permitting – link
- [10] GREENHOUSE GAS EMISSIONS FROM AI (ICEF) – link
- [11] Knight Frank – Data Centres: Taking Stock of Sustainability – link
- [12] Carbon Brief – AI: Five charts that put data-centre energy use and emissions into context – link
- [13] 50 GW of datacenter demand queues up for UK grid access – link
- [14] Rapid growth in number of proposed data centres raises huge concerns for UK climate targets – link
- [15] Data centres must be included in UK’s decarbonisation plans – link
- [16] Data Centre Impact Study – National Grid DSO – link
- [17] New UK AI datacentre could cause five times emissions of Birmingham airport – link
- [18] Datacentre developers face calls to disclose effect on UK’s net emissions – link
- [19] Will increasing data centres undermine UK net-zero commitment? – link
- [20] Planning for data centres in 2026 – What’s changed and what will matter most (news/insights) – link
- [21] Data Centre Programme – What we achieved in 2025 – techUK – link
- [22] RWE agrees 10-year PPA with data centre provider – link
- [23] What Are Power Purchase Agreements (PPAs) in the Context of Renewable Energy for Data Centers? – link
- [24] Power Purchase Agreements (PPAs) for AI Data Centers – Introl – link
- [25] Everything data center operators need to know about Power Purchase Agreements (PPAs) – link
- [26] The renewable route for data centre expansion – Carbon Trust – link
- [27] Manchester Data Centre to be First in the UK to Install Hydrogen-Ready Fuel Cell –
If this connects with something you are working on, send me a note. I am interested in serious conversations around hydrogen, batteries, infrastructure, advanced materials and deep tech commercialisation.
