Commentary. Artificial intelligence is being sold as both an economic engine and a possible climate tool. It may become either—or neither—depending on whether governments treat its energy use as a public-interest question rather than a private-sector footnote.
The issue is moving quickly. AI systems require large data centres, and those facilities consume substantial amounts of electricity and water. The carbon consequences are becoming prominent enough that climate officials in Turkey, host of next month’s COP31 summit, say they intend to promote “climate-responsible AI.” Reporting on the debate notes that AI’s environmental footprint is expected to receive heightened attention at the summit.
Our view is straightforward: governments should stop granting AI infrastructure an automatic presumption of social value. Companies seeking public subsidies, expedited permits or access to scarce electricity should have to show what their projects contribute—and how they will limit the damage.
The case for restraint
The concern is not technological pessimism. It is accountability. Data-centre construction can lock regions into decades of additional electricity demand just as governments are trying to electrify transport, heating and industry. In places where grids remain dependent on fossil fuels, more computing can mean more emissions, regardless of how innovative the software appears.
There is also a distributional question. The benefits of AI investment may accrue to a small number of companies and highly paid workers, while the costs—higher grid investment, pressure on water supplies and potentially increased utility bills—are shared by the public. That imbalance is especially difficult to justify when officials describe AI as a national priority without publishing clear projections of its local impacts.
The broader economic backdrop strengthens the case for caution. The International Monetary Fund has warned that an AI-market correction is among the risks facing Asia’s outlook, alongside energy shocks and higher global borrowing costs. Coverage of the IMF meetings also describes growing credit pressure in the data-centre sector, where projects can depend heavily on expected future revenue.
The argument on the other side
There are serious counterarguments. AI could improve electricity forecasting, identify methane leaks, optimise industrial processes and accelerate scientific research. Better models may help utilities integrate intermittent renewable power or reduce waste in logistics and manufacturing. Blocking new infrastructure indiscriminately could also push investment to jurisdictions with weaker environmental standards.
Nor should every kilowatt-hour used by a data centre be treated as waste. Digital services have become part of modern economic and scientific infrastructure, and communities may reasonably seek the jobs, tax base and investment that new facilities bring. Excessively rigid rules could favour established technology giants over smaller firms that cannot absorb compliance costs.
These points deserve more than dismissal. The answer is not a blanket ban or a symbolic cap on computing. It is a system that distinguishes demonstrable public benefit from marketing language.
What responsible policy would require
First, companies should disclose energy and water use by facility, along with the carbon intensity of the electricity they consume. Second, public authorities should require credible plans for using new clean power rather than allowing firms to claim renewable energy through accounting instruments alone. Third, approvals should include enforceable efficiency targets and emergency limits during grid stress.
Most importantly, governments should measure outcomes. If an AI system claims to cut emissions, the claim should be tested against a baseline and reviewed independently. A model that saves energy in one application but encourages far greater consumption elsewhere should not receive an unqualified climate label.
This approach would not eliminate AI’s environmental cost. It would make that cost visible—and give citizens a say in who pays it. The technology may indeed become a useful climate tool, but that possibility cannot serve as a blank cheque for expansion today.
At COP31 and beyond, the right question is not whether AI is good or bad for the planet. It is whether its owners, regulators and beneficiaries are prepared to prove that the benefits justify the bill.