Electricity demand from data centres is set to roughly double by 2030 in Britain. If forecasts are correct, they will consume almost as much electricity as the whole of Wales does today. It is easy to see why this isn’t an exciting prospect to the already-stretched lay bill payer. Yet data centres may be a blessing in disguise for the grid. Let me explain.
Britain is building a grid for a world where electricity demand, pardon the pun, surges. The problem is every forecast of rising electricity use has turned out to be wrong. High electricity prices have created a massive incentive to cut back. Consumers have switched to energy efficient appliances, while heavy industry is in painful decline. Demand for energy-saving electric technologies such as heat pumps and EVs has been weaker than hoped for. Since 2000, electricity use has fallen by 16%. When you control for population growth, it is even more stark. Since 2000, Britain had the largest per-capita drop in electricity consumption of any OECD nation.
In many markets, a fall in demand leads to lower prices, but electricity markets do not work like other markets, at least not in the short-term. Every electricity bill is a bundle of variable and fixed costs. Less demand may mean using less gas (and in turn, not having to buy from the most expensive sources), but fixed costs like the grid remain the same. The overall bill may fall, but the price per watt increases. As Britain moves away from fossil fuels and switches to a mostly renewable system, fixed costs will make up a bigger and bigger share of the bill.
Think of the electricity market like a fast food restaurant. The owner has paid rent on a new premises and bought a state-of-the-art deep-fat fryer. Potatoes are cheap, but the cost of a portion of chips reflects not just the price of labour and raw ingredients, but also the need to cover the upfront cost of the premises and equipment. The more customers buying fries the more diluted the upfront costs become. Busy restaurants can sometimes afford to charge less.
A relatively small fall in consumption can have big implications for cost. One study from Boston Consulting Group estimated that the 11% fall in electricity demand since 2015 is responsible for almost two-fifths of the increase in system costs. System costs are now so high that – assuming demand stays the same – bills are forecast to slightly increase in 2030 even if the wholesale price of electricity collapses.
This isn’t a new insight for markets like energy. In fact, Britain’s grid was built under it. As energy historian Arthur Downing notes, the Central Electricity Board (CEB) had a clear plan from day one: “drive down prices, drive up consumption, spread the fixed costs and build infrastructure to drive down prices.” In other words, the opposite of current policy. The CEB and its post-war successor, the Central Electricity Generating Board, were, in Downing’s words, ‘load-hungry utilities’. And the hunt for ‘load’, or demand spread evenly across the day/year, delivered large benefits for consumers. Between the 50s and 70s, electricity demand grew fourfold while prices fell by more than a fifth in real terms.
Gas, Downing notes, took an even more extreme approach actively bundling gas and gas equipment.
“In the 1890s gas utilities realised they could never compete with electricity for lighting in upper- and middle-class homes, and so went after other types of load. The industry worked with equipment manufacturers to sell gas-fired cookers and heaters to affluent homes. At the other end of the market, gas utilities focused on selling working-class households a package of Welsbach gas-mantle lights and a prepayment meter. Gas utilities also pioneered financing models in which equipment, connection and commodity costs were bundled into a single unit rate.”
Rediscovering this insight is arguably more important now than ever before as we try to shift to clean homegrown power. System costs are growing.
Wind and solar are, in many ways, fantastic technologies, but integrating them onto the grid is expensive. Most of our wind resource, particularly offshore wind, is in Scotland and the North while most demand is in the South East. As it stands, Britain’s grid isn’t up to the task of moving power from where it’s generated to where it’s needed. Renewables also require a degree of overbuilding to meet demand when it isn’t very sunny or windy. The end result is that wind farms are increasingly being paid to switch off (while gas plants are being paid to fire up at extremely short notice). While nuclear has its own problems: it struggles to compete with clean renewables when it’s windy and sunny, and dirty gas when it isn’t. The big upfront costs of nuclear construction are shifted onto billpayers as a fixed cost as a result.
How data centres can help dilute system costs
In the past, Britain’s grid builders cut costs by growing a diversity of demand. While adding to peak demand necessitates greater build out, demand outside of peak times effectively dilutes fixed costs. One modern-day equivalent is air conditioning. AC is useful when it is hot and sunny, or put differently, when there’s little demand for heating and plenty of solar generation. That’s one of the reasons why Britain Remade (and others) have argued that rules restricting AC uptake should be removed.
Another way to dilute system costs is to add consistent reliable demand to the grid. Data centres do this better than any other plausible source of new power demand. They consume electricity at a relatively consistent rate. Unlike factories that operate in shifts, demand doesn’t dip at weekends or overnight.
Most industrial demand benefits from targeted support. Heavy industry typically qualifies for the British Industry Supercharger relief, which exempts them from many of the policy costs on an energy bill. This relief is itself funded by higher charges on every other bill. There’s a strong case for this approach – after all, we want to avoid losing even more heavy power customers. Yet the Supercharge relief doesn’t apply to data centres. In other words, when data centres pay for power, they’re covering a fair greater share of fixed system costs than other big customers.
What’s the prize? Energy analyst Robert Boswall estimates that hitting the 6GW UK Compute Roadmap target would save consumers around £2bn each year.
Data centres can do even more
An annual saving of £2bn is welcome, but at just £69 per household it won’t change the fundamental fact that Britain’s electricity bills are among the developed world’s highest. Dilution, however, isn’t the only (or the biggest) opportunity for data centres to cut bills for the rest of us.
Britain’s energy intensive businesses are struggling due to high prices. Politicians may speak about reindustrialisation, but the painful reality is that Britain is deindustrialising at an unprecedented rate. Since 2021, Britain has lost a third of its energy-intensive businesses. Cement, chemicals, and ceramics have all seen dramatic declines. Household names like Denby have been forced into administration.
High power prices matter far less for data centres. This isn’t because energy isn’t a large chunk of the cost of a data centre – it is. Epoch AI estimates that energy represents 7% of the total cost of building and operating a 1GW data centre. Their figure is based on the US, where electricity is far cheaper. At UK prices, electricity rises to about a fifth of the all-in cost. Data centres aren’t quite as electricity-hungry as aluminium smelting (30–40% of costs) or chlorine production (around half of costs), but that still puts them on a par with electric arc furnace steelmaking .
Where data centres differ from other energy-intensive businesses is in the value they gain from each megawatt hour. Most of that value is generated from highly-sophisticated chips and the ideas of super-smart engineers.
AI inference is extremely useful and businesses are willing to pay a lot for it. Anthropic has generated £25Bn in revenue so far this year. SemiAnalysis, an AI research firm, estimates that API inference can generate around $100bn per 1GW data centre. Even at the UK’s ultra-high electricity costs, power is just 3% of revenue.
There’s another way of thinking about it. At $100bn per GW per year, OpenAI could buy Hinkley Point C, which when finished will be the most expensive nuclear power station in history, at full price and it would take just 68 days to pay it off.
The problem for firms like Anthropic and OpenAI isn’t high power prices – though they don’t help – it is the massive waits for access to the grid. Microsoft recently pulled out of plans to build a £2bn data centre, the UK’s largest, in Essex after learning its grid connection has been delayed to the 2030s. Speed is everything.
Britain’s grid operates on a first-come, first-served basis. There are few fees attached to entering and progressing through the grid queue. When a business gets to the front, they only have to pay a fraction of the full costs of connection – while things like the cost of upstream reinforcement go onto household bills. Access to the grid is extremely valuable – projects cannot proceed without it – but businesses are charged only a tiny fraction of what it is worth.
Businesses respond to incentives. When valuable resources are under-priced, demand swiftly outstrips supply. The end result is access to the grid is allocated on the basis of willingness to queue, not willingness to pay. Waits for grid connections can be upwards of 10 years. Many applications in the grid queue are speculative. They are not serious projects, but rather a safe bet that a place near the front will be valuable to someone else. A data centre that could generate billions in revenue ends up stuck waiting behind applications for projects that may never be built.
The Government’s solution is to police the queue better. And move to a system of first-ready, first-served. It’s what I call the ‘soup nazi’ solution. In one episode of the sitcom Seinfeld, there’s an extremely popular soup restaurant. Queues are long but move fast because customers who get to the front without deciding their order are told “no soup for you” and made to leave.
This is an improvement on the status quo, but not much. Something extremely valuable is still given away for next-to-nothing. There is no real priortisation based on who values a fast connection the most. A better approach would let the most valuable projects pay to jump the queue. In essence, AI data centres should be able to buy a Thorpe Park style fast-pass.
To connect the grid queue will be expensive. It will require bigger substations, transformers, and new high voltage lines. That’s currently paid by network charges that go on every single household bill, even though some benefit from it far more than others.
A better approach would copy the approach Britain has taken to the electromagnetic spectrum. In the year 2000, Britain ditched its historic approach of allocating the electromagnetic spectrum by ‘beauty contests’ (bureaucrats reviewing applications) and instead decided to auction off the 3G bands to whoever would pay the most. It raised £22.5bn (equivalent to 2.5% of GDP).
Auctioning off grid connections could be just as lucrative. The option to skip the queue by just a few years would be extremely valuable to some firms. Boswall, who advocates this approach, estimates that developers would willingly pay £1.3m per MW per year of accelerated access. If that is the case, then accelerating 3GW by 3 years would be worth £12bn. This is a conservative estimate: Boswall’s figures come from established operators like Vantage, and frontier AI firms racing for compute would likely pay more.
There is currently 50GW of data centre capacity within the grid queue. Money raised from auctions could go to taking some network charges off household bills. This wouldn’t just save money, it would also make it much more attractive for consumers to switch to green electrotech like heat pumps and EVs.
Bring your own power
There is another way to use the high value data centre developers place on access to the grid to benefit us all: get them to build their own power.
Instead of waiting for grid connections, developers could build on-site (or nearby) generation then connect it directly. In the short term, gas is likely to be the best option. In the US, data centres have placed orders for 75GW of behind-the-meter generation capacity. For context, that’s more than double Britain’s entire gas fleet. About a quarter of the new generation in the US is on-site behind-the-meter power for data centres. The US has similar, though not as extreme, grid connection problems.
The big challenge is making all that power clean. Most of the power added behind-the-meter is gas. Without access to the grid, there is a premium on reliability. Gas meets that requirement. Nuclear does too, but takes far longer to build. Some tech companies (including Microsoft) are paying to restart and extend the operating life of existing nuclear plants. Renewables are more challenging. Solar and wind may be quick to build, but they are relatively land-hungry and have low load-factors (they only run when it’s sunny or windy). Without access to the grid for stability, they are unlikely to be a solution behind-the-meter. There is a risk that in the short-term meeting the demand for data centres raises emissions.
With the right policies, however, data centres could still be useful in the rollout of clean power. One option would be to allow data centres to connect to the grid flexibly. In other words, they could get a grid connection ahead of schedule provided they agree to switch off (or use their own power) near peak times. This matters because much of the delay in getting a grid connection is less about day-to-day use, but dealing with extreme peaks. One study looking at the US’s largest regional grid, PJM, found that it was possible to add 13GW of new load without major new grid capacity upgrades provided that the new load is flexible for an extremely small (0.25%) share of its uptime. In other words, when demand is close to peak they must be ready to switch off. Flexible connections do exist, but they tend to be set up for generators and batteries. They also typically require about ten to twenty times more curtailment than the US study.
Britain may be moving towards a ‘clean’ grid, but gas will play an important role for a long time. Our aging gas fleet will need to be replaced at some point. One option would be to allow data centres to use gas as a bridge to clean power. In five to ten years, data centres could transition to on-site small modular nuclear reactors or use power-purchase agreements to buy the same equivalent power from renewables. In the US, there are already deals being struck between tech companies like Google and Amazon and SMR-builders like X-Energy and Kairos.
Finding the right model for this will be challenging, but data-centres need power and are willing to pay a lot for it. We can get a lot built if we let them.
Very interesting. How can government be persuaded to implement the model you have suggested which makes sense to me and gets the private sector to pay for much of the infrastructure.
The same model could be adapted to the road and rail sectors.
My first post uni job in 1990 was about expanding gas production in Australia as a transition fuel before renewables proper would kick in. I think over 36 years is more than enough to be considering gas a transition fuel. We are now past that point and we just need to say no to any new fossil fuel demand. There is already too much CO2 in the atmosphere whose effects are causing lots of problems.