On 9 November, Prime Minister Andy Burnham identified AI as a priority for Britain's 2027 G20 presidency. Yet in the US elections days earlier, voters favoured limits on data-centre development in 40 of 41 local ballot contests tracked by Ballotpedia. The mix of bans, restrictions and advisory votes highlights the tension between national AI ambitions and local willingness to host its infrastructure.
At COP31, this tension raises a practical question: how can we build the computing capacity needed for scientific discovery and better services while cutting emissions and securing public support?
Who bears the costs
The International Energy Agency's central projection sees global data-centre electricity use almost doubling, from 485 terawatt-hours in 2025 to around 950 in 2030. These figures cover all data centres, not AI alone. When demand grows faster than clean power and grid connections, securing electricity quickly can undermine climate commitments.
A study in Nature Climate Change links data centres built after 2020 with nearby coal-fired generating units remaining open longer. This is an association, not proof of causation. Meanwhile, US research in Nature Energy finds that wholesale-price pressures and rules for sharing grid costs can leave rural and lower-income households disproportionately exposed. The effects depend on local systems and policy choices.
For neighbours, the questions are immediate: will bills rise, will water remain available during droughts, and will promised jobs justify construction disruption, noise and land use? Benefits may reach users worldwide while pressures fall on particular communities. That uneven distribution makes data centres a key part of the COP debate about a fair transition.
Why the opportunity matters
The case for AI extends far beyond convenient answers. It can help researchers investigate disease and identify promising materials for batteries and other clean technologies. Better weather forecasts can help communities prepare for extremes, while AI-based flood warnings can support climate adaptation where monitoring is limited. Within energy systems, AI can improve renewable-generation forecasts, support grid operations and reduce waste in buildings and industry. These uses connect the same infrastructure that draws on local resources to urgent challenges in health, energy security and climate adaptation.
These are substantial opportunities, but their value depends on reliable applications reaching those who need them. Access to computing and skills matters alongside investment in facilities. Nor do potential benefits automatically cancel out environmental costs: more efficient hardware and software can reduce energy per task while wider adoption increases total consumption.
Engineering for local conditions
Engineering can ease these pressures. Smaller models, where reliable, and fewer idle servers reduce unnecessary demand. Some computing tasks can move to times or places with more clean electricity and spare grid capacity. Immediate-response services and costly interruptions to training limit this flexibility.
New facilities should be planned alongside additional clean generation, storage and grid capacity. Annual renewable-electricity purchases alone cannot establish that clean power is available where and when it is needed.
Cooling choices must reflect local conditions. Treated wastewater can reduce reliance on drinking water. Recirculating coolant and releasing heat without evaporation can cut on-site water consumption, although mechanical cooling may require more electricity. Water used in power generation and manufacturing still counts. Waste heat can warm nearby buildings where suitable networks and demand exist; the benefit depends on the heating displaced and the electricity required.
More experimental options include solar-powered computing in orbit. Google's Project Suncatcher has reported prototype AI computation in space, suggesting a possible way to ease terrestrial resource pressures. Its environmental value remains unproven: launches, manufacturing, hardware replacement, disposal and radiating waste heat all present challenges.
Earning public support
Better engineering gives communities more credible options, but it cannot decide who should pay or which developments should proceed. Grid agreements should assign costs fairly and protect households if projected demand never materialises. Developers also need predictable requirements to make investments that can take years to deliver. Communities need a meaningful say over whether, where and on what terms projects proceed, with enforceable commitments on water, noise and local benefits.
Public reporting should show absolute electricity and water use, supply-chain emissions and benefits actually delivered, including lasting employment and useful heat supplied to neighbouring buildings. This lets residents and policymakers judge whether a project merits support. Opposition may lead to better proposals, delay construction or move pressures elsewhere; treating it simply as an obstacle to investment misses the choices at stake.
The challenge is to connect AI's wider promise with conditions people can trust where its infrastructure is built. Clean power, fair cost sharing and meaningful public influence should develop alongside computing capacity. This is where COP31 comes in, where political ambition can translate into lasting public benefit.