Who pays for AI? Data centres, electricity costs and regional energy equity

As data centres proliferate across the globe, underpinning the AI boom, the central issue for policymakers is now shifting from how to power these centres, to who pays for the grid capacity, electricity generation and backup supply they demand. Yanxi Zhou and Tao Zou explore how concerns around the costs of data centres and their fair distribution should be addressed, drawing on examples from China, the UK and US.

Britain’s VAT cut and the hidden cost of the data centre boom

When UK Prime Minister Andy Burnham recently announced that value added tax (VAT) on household electricity bills would be reduced from 5% to zero from 1 October 2026, the policy was framed as immediate relief on winter bills. However, the VAT cut coincides with a deeper shift: AI data centres are reshaping electricity systems and demand worldwide. In early 2025, the UK government estimated that data centre electricity consumption in the UK will increase fourfold by 2030. To understand why this matters, we must look beyond the electricity costs and into the machinery that sets prices in the first place.

The scale of the new demand from data centres

Across the globe, data centres today consume around 415 terawatt-hours (TWh) of electricity annually, about 1.5% of global electricity demand. This could more than double to 945 TWh by 2030 under the base case. AI-optimised facilities are expected to account for much of the increase, with accelerated servers responsible for nearly half of the net growth.

The supply pressure is concentrated in particular countries, with China and the US already accounting for more than half of global data centre electricity use. Data centre electricity use in the US could reach 325–580 TWh by 2028 and will account for 20% of total US electricity demand by the end of the 2030s. Meanwhile, China’s computing-related demand could approach 800 TWh by 2030, around 6% of national electricity use. In the UK, data centres currently consume about 2.5% of grid electricity, but proposed AI-driven projects could raise this to 8.8% of national demand by 2030, requiring up to 50 gigawatts (GW) more than the country’s current peak demand.

Overall, the International Energy Agency (IEA) expects data centres to drive almost half of global electricity demand growth this decade. A more aggressive estimate suggests global data centre demand could reach nearly 1,935 TWh by 2033, while US facilities could consume up to a fifth of national generation by 2035, mostly driven by training and inference workloads. These figures underline that the huge expansion in AI use and applications is becoming a major challenge for national electricity systems.

Why the grid is starting to charge everyone else

When a large new load such as an AI data centre arrives on an electricity grid faster than local generation or transmission capacity can expand, it can push up wholesale prices for everyone drawing from that grid. It is estimated that existing US data centres have already lifted wholesale electricity prices nationally by 2% to 6% on average, with sharper effects in grid regions most exposed to AI activity: AI demand significantly increases wholesale electricity prices by up to 25% in American Electric Power zones of the PJM grid, such as Virginia. Under a scenario when most data centres get built and run at high capacity, wholesale electricity prices could climb by around 50% by 2028.

Three mechanisms explain this surge in electricity prices:

• Shared infrastructure: If access to new electricity substations, transmission lines and reserve capacity is shared across all customers, households will effectively subsidise facilities built mainly for private data centres.
• Peak generation: Utilities rely more on expensive ‘peaker’ power plants, which are designed to run only when demand is high, raising wholesale and retail prices.
• Delayed decarbonisation: Overloaded grids may keep older coal and other high-emission power plants operating for longer, increasing environmental costs and slowing the clean energy transition.

Two different tools for two different problems

A VAT cut is a demand-side relief measure: it is quick, visible and temporary. But it barely addresses the underlying costs of electricity generation, transmission, distribution and grid expansion associated with demands from data centres.

Pressures on electricity prices driven by AI require a different approach: cost allocation. The UK’s energy regulator, Ofgem has proposed a commitment fee of 2.5% to 7.5% of average data centre project costs to secure connection to the grid. It is also considering whether developers should finance and build their own grid access. In the US, Virginia has introduced a 1.1% per kilowatt hour data centre charge, while North Carolina has withdrawn a power tax exemption for data centres. These developments signal that governments are re-examining how electricity is subsidised as the scale of AI data centres’ draw on the grid becomes clear.

Other approaches to address these rising prices include commitments from companies themselves and stricter rules on grid connection. Major technology companies, such as Google and Meta, have signed the Ratepayer Protection Pledge to cover new generation costs. Ireland, a significant data centre hub, requires new centres to provide dispatchable power or storage (which can respond at the request of grid operators), and source at least 80% of annual demand from domestic renewables.

The common principle in these approaches is that the entity causing the new cost will cover it, through dedicated large-load tariffs. In this way, they aim to address the fact that new transmission lines, substations and reserve capacity are built largely to serve a small number of large private customers, while often paid for by everyone on the grid.

How to relieve the increasing electricity price pressure for communities

One immediate solution to the increasing electricity prices is introducing fair cost allocation. Under this scenario, data centre operators pay for the grid capacity, storage, backup supply and infrastructure they need. Another solution is to reduce the cost per watt of electricity or control the rising speed of electricity demand from data centres. These relief measures need to address the structural drivers of price inflation and effectively socialise those costs across all ratepayers.

Data centres can also reduce their dependence on public grids by procuring or producing clean power directly, so that new AI load does not feed into existing grid users’ costs. Examples include Google’s 15-year, 100 megawatt, offshore-wind power purchase agreement for European AI growth; and Microsoft’s collaboration with Nvidia on advanced nuclear energy. At the household level, local microgrids using solar power and batteries can reduce exposure to peak prices, though their benefits depend on affordable equipment and tariffs that reward local generation.

Finally, AI’s electricity demands can be reduced through greater efficiency, ensuring that we can create capacity and support innovation without making communities absorb the costs. Improved efficiency could be achieved through specialised chips, power caps on graphics processing units , liquid cooling, smaller models for simpler tasks, local inference and AI-assisted grid scheduling. Google’s Trillium TPU, for example, reportedly delivers 4.7 times more compute, while using 67% less energy.

Regional equity dilemmas for infrastructure hosts

For regions hosting key AI infrastructure, such as large data centres, there is a need to balance three competing priorities: technological and economic progress, environmental sustainability, and equitable distribution of benefits and burdens.   

In the UK, South Wales is projected to host 54% of National Grid Electricity Distribution’s additional capacity by 2050, driven partly by the Vantage Data Centres campus near Newport and the UK Government’s AI Growth Zone. While host regions may gain investment and jobs, they may also face higher electricity demands, grid upgrade costs, pressures on land and water use, and impacts on household bills.

China has a similar challenge. Its East Data, West Computing strategy sought to move computing capacity towards western regions with abundant, cheaper power. Yet rising data centre electricity prices in Beijing suggest that computing is being pulled back towards eastern demand centres, where users and infrastructure are concentrated.

These examples show that if infrastructure expansion costs pass through to all ratepayers, residents in host regions will end up subsidising a digital service consumed far beyond their borders, raising equity concerns around who should bear these costs. That risk is sharper in emerging economies, where clean energy finance is limited, and many households remain underserved of energy. In these areas, data centres could crowd out residential access, industrial electrification and grid expansion.

Who pays the AI energy bill: fragmented approaches and the need for public protection

In the US, AI demand on electricity grids is leading to wholesale price shocks and is being dealt with by fragmented state-level responses: Virginia has introduced new taxes on data centres, while North Carolina has pulled back its tax exemption. The UK has responded to high household electricity bills with a tax cut, while not addressing longer-term costs of data-centre expansion. China has, so far, absorbed pressure through surplus renewable capacity, but faces renewed strain as computing demand grows and clusters closer to population centres. However, the stakes are highest in lower-income, capital-constrained regions hosting AI infrastructure, such as Malaysia’s Johor state. Public incentives need to be tied to demonstrable local benefits and protections on affordability.

Every electricity system around the world now faces the same test: how to divide the cost of powering AI among the companies driving demand, the households sharing the grid and the regions hosting the infrastructure. And alongside this, improving data centre and grid efficiency, to ensure electricity use grows predictably, without destabilising prices or limiting household access.