Gen AI is the grid's next big challenge. Is now the right time?

1. The scare has good numbers

The demand is real, big, and concentrated in the wrong places.

Let's give the worriers their due, because their numbers are genuine. NESO counts around 140 data centre proposals seeking roughly 50GW of grid connections — about the same as the whole of Britain's peak demand.¹ Its central view has some 5.2GW of that actually connected by 2030, drawing just over 20TWh a year, and its Future Energy Scenarios have data centre demand rising from 7.6TWh in 2024 to somewhere between 20 and 41TWh by 2035 — a spread so wide it's really a shrug with error bars.² The government, meanwhile, wants AI Growth Zones of at least 500MW each, with at least one scaling past a gigawatt.³

And the location problem is real. Around two-thirds of existing UK data centres sit within twenty miles of London, on the most congested part of the network — famously, new housing in west London stalled for years because data centres had claimed the spare capacity first.¹ So yes: big new load, lumpy, badly placed, arriving fast. If you stop reading there, the concern writes itself.

2. Look what else just happened

The supply side and the plumbing both unblocked in the same eighteen months.

Here's what the worried take skips. While the data centre queue was forming, two of the longest-standing blockages in British energy quietly cleared.

The first is onshore wind. England spent nine years under a de facto ban — since 2015, a single local objection could kill a project — and that ban was lifted in July 2024. A year later came the first-ever Onshore Wind Strategy, forty-odd actions aimed at nearly doubling capacity to as much as 29GW by 2030.⁴ The cheapest new generation Britain can build is legal again, at precisely the moment a wall of new demand shows up asking for it.

The second is duller and, I'd argue, matters more. Market-wide Half-Hourly Settlement — the reform that finally settles every meter in the country on the half hour rather than on an estimate — started migrating meters in October 2025, expects roughly 80% across by October this year, and completes in 2027, cutting the settlement cycle from fourteen months to four.⁵ That's the plumbing that lets suppliers see demand as it actually happens, price it, and pay people to move it. You cannot run a flexible grid on estimated meter readings, and until now, mostly, we did.

So the timing question inverts. The concern assumes the demand is arriving at the worst possible moment. On the evidence, it's arriving at the first moment in decades the system could actually do something useful with it.

3. Demand is the point, not the problem

A wind farm's hardest problem isn't planning. It's a buyer.

Renewables don't get built on targets; they get built on revenue certainty. What a developer needs to reach a final investment decision is a creditworthy buyer signed up for fifteen years — and there is no better anchor tenant on earth right now than a hyperscaler with a board-level clean energy commitment. This isn't hypothetical. Google has a stated goal of running on carbon-free energy every hour of every day by 2030 and already matches around 74% of its consumption hourly; Microsoft signed a twenty-year deal to restart an entire nuclear plant at Three Mile Island for its data centres.⁶

Notice what that second example implies. These buyers don't want annual certificates that net off over a year — the old trick where you buy Scottish wind on paper and burn gas at 6pm. They're contracting for around-the-clock clean power, which forces the portfolio to be designed properly from day one: wind plus solar plus storage plus something firm, shaped to a real load. That's the "from the ground up" system Britain keeps writing strategy documents about, except this time a customer is paying for it rather than a subsidy scheme coaxing it along.⁷

4. The homework we keep not doing

Decoupling gas from the power price has been "due" since 2022. Nothing was due tomorrow, so nothing got done.

Now for the part I actually care about. Britain's electricity price is set by the most expensive generator running in each half hour, and that generator is nearly always gas — gas set the price in as many as 98% of hours in 2021, and 84% of the time across recent years, despite producing well under half our electricity.⁸ That's why the 2022 gas crisis walked straight into everyone's electricity bill, and why "decouple electricity from gas" has been on the national to-do list ever since. The Review of Electricity Market Arrangements ran for three years, considered every option up to and including tearing the market into zones, and settled in July 2025 on reforming the national price instead.⁹

Which is to say: we did what we always do. We bought the highlighters, made the revision timetable, and put the essay off. It's the British way — nobody in this country has ever started their homework before the night before, and honestly, some of our finest work has come out of that panic. The trouble is that decoupling has no exam date. Clean Power 2030 sets a target of 95% low-carbon generation, and NESO calls it achievable without raising costs,¹⁰ but a target a government sets itself is a deadline only in the sense that a gym membership is a fitness regime.

Here's the nuance that makes the gas problem harder than it looks, and the data centres more useful than they look. Even at 5% of generation, gas can still set the price — being rare doesn't stop you being the marginal unit at the margin. Gas stops setting the price only when something else can profitably be the flexible thing in the tight hours: storage, demand that moves itself, firm clean capacity. And that is exactly what a 24/7 clean-power buyer pays to build. Round-the-clock contracts fund the batteries and the flexibility directly, because hourly matching is the product being sold. The data centre queue isn't a threat to the decoupling project. It's the first customer the decoupling project has ever had.

5. The night before the deadline

So, is now the right time? Wrong question, I think. There was never going to be a right time — there was only ever going to be a forcing event, because that's how this country operates. The demand is enormous and awkward and it's going to strain the grid, and it's also underwritten by the deepest-pocketed, cleanest-committed buyers the electricity market has ever seen, landing just as onshore wind comes back and half-hourly settlement makes flexibility a real market. The concerns about data centres aren't wrong on the facts. They're wrong on the framing. The essay question was set in 2022. The exam is finally tomorrow morning. And on past form, that's the only condition under which Britain has ever produced its best work.

Notes & sources

1. NESO counts ~140 data centre proposals seeking around 50GW of grid connections, roughly equal to GB peak demand; some two-thirds of existing data centres sit within 20 miles of London, and spare capacity claims there famously stalled new housing in west London. Drax Electric Insights Q1 2026, "Britain's AI ambitions depend on the grid". ↩
2. NESO's central analysis points to ~5.2GW of connected data centre capacity and just over 20TWh of demand by 2030, with connection requests far exceeding that; Future Energy Scenarios 2025 models data centre demand rising from 7.6TWh (2024) to 20–41TWh by 2035. Data Center Dynamics, on NESO's forecasts. ↩
3. The UK Compute Roadmap targets AI Growth Zones each serving at least 500MW of demand by 2030, with at least one scaling beyond 1GW. Written evidence to Parliament, DCU0081. ↩
4. England's de facto onshore wind ban (in place since 2015) was lifted in July 2024 via reform of the National Planning Policy Framework; the first Onshore Wind Strategy, published 4 July 2025, sets out 40+ actions to reach up to 29GW by 2030. Osborne Clarke, on lifting the ban; Trowers & Hamlins, on the strategy. ↩
5. MHHS migration began in October 2025, with ~80% of meters expected to be migrated by October 2026 and completion in May 2027; full half-hourly settlement cuts the settlement cycle from 14 months to four. Elexon, Market-wide Half-Hourly Settlement. ↩
6. Google targets 24/7 carbon-free energy on every grid it operates on by 2030 and reports ~74% hourly matching worldwide; Microsoft signed a 20-year agreement with Constellation to restart Three Mile Island Unit 1 (835MW) exclusively for its data centres. McKinsey, on the race for 24/7 clean power. ↩
7. 24/7 clean-power PPAs contract hourly-matched portfolios blending wind, solar, storage and firm low-carbon sources, moving beyond annually-netted renewable certificates. Data Center Dynamics, on the 24/7 CFE marketplace. ↩
8. UCL research found gas set GB electricity costs 84% of the time in recent years despite generating under half of supply; a widely cited analysis put the figure at ~97–98% of hours in 2021. UCL, on gas setting electricity prices; Carbon Brief, on marginal pricing. ↩
9. The REMA Summer Update of 10 July 2025 rejected zonal pricing and committed to a reformed national pricing model, retaining a single GB-wide wholesale market. Norton Rose Fulbright, REMA Summer Update. ↩
10. The Clean Power 2030 Action Plan targets at least 95% of generation from low-carbon sources with no more than 5% unabated gas; NESO's advice judged this achievable without increasing costs and said it would insulate the UK from volatile international gas prices. DESNZ, Clean Power 2030 Action Plan; Carbon Brief analysis. ↩

← All projects