Commentary. Artificial intelligence is increasingly being treated as both an economic strategy and a diplomatic priority. That is understandable: the technology could improve productivity, scientific research and public services. But governments are moving faster to attract data centres than to settle the basic question of who pays for the electricity, water and infrastructure they require. Our view is straightforward: AI expansion should proceed, but only under enforceable public-interest conditions.
The issue is no longer theoretical. G7 foreign ministers meeting in Paris are discussing artificial intelligence alongside Iran, Ukraine and energy security, a sign that the technology has moved from a niche industry concern into the centre of international policy.[1] The United States has warned that international regulation could stifle innovation, while still acknowledging that rapidly developing AI creates serious risks.[2] That argument deserves to be taken seriously. Excessively rigid rules can protect incumbent firms, slow useful research and leave democratic countries dependent on less transparent competitors.
Yet “innovation” cannot be a blank cheque. The European Union’s proposed framework for strengthening its cloud and AI ecosystem explicitly calls for data-centre technologies to incorporate energy and resource efficiency by design and throughout their operations.[3] That principle should become the baseline everywhere, not an optional aspiration attached to subsidies.
The bill is larger than the server farm
Data centres consume substantial amounts of electricity and water, particularly where cooling systems rely on scarce local supplies. Reporting ahead of the United Nations climate summit in Antalya cited projections that data-centre water use could reach 73 billion litres a year by 2028.[4] Such estimates vary by facility, cooling technology and accounting method, but the direction is clear: the physical footprint of digital services is growing just as governments are asking societies to conserve resources.
This creates a democratic problem. A new facility may bring construction jobs, tax revenue and investment. It may also raise demand on grids, accelerate transmission projects and compete with households or existing industries for power. If the costs are socialized while the gains accrue mainly to technology companies and investors, public opposition should not be dismissed as anti-progress. It may be a rational response to an unfair bargain.
There is a legitimate counterargument. Delaying projects could push investment elsewhere, weaken national competitiveness and deprive smaller cities of badly needed economic opportunities. AI systems may also produce efficiency gains that reduce emissions in transport, manufacturing and energy management. Those benefits are plausible, but they are not automatic. A claim that future efficiencies will compensate for present resource use must be demonstrated, measured and publicly audited.
Conditions, not blanket bans
The answer is neither a moratorium on data centres nor a race to approve them. Governments should require developers to disclose expected electricity and water use, finance grid upgrades tied to their demand, and meet independently verified clean-energy and efficiency targets. New facilities should not receive preferential power contracts during shortages unless they can show a compelling public benefit. Local authorities should have meaningful authority over siting, water withdrawals and emergency plans.
Those rules would not eliminate every dispute. They would, however, make the dispute more honest. Companies could compete on engineering and efficiency rather than on promises of jobs alone. Residents could judge projects using comparable information. Policymakers could support innovation without pretending that computational growth has no material consequences.
The same logic applies internationally. At COP31 in Türkiye, nearly 200 countries are negotiating over climate finance, clean energy and adaptation while communities already face mounting climate risks.[5] It would be incoherent for governments to demand faster decarbonization from households and poorer countries while granting high-energy digital infrastructure an informal exemption.
AI policy is often framed as a choice between regulation and progress. That is a false choice. The real choice is whether progress will be governed early enough to retain public trust. Our newsroom believes the sector should be welcomed—but on terms that protect communities, price environmental costs honestly and make measurable public benefit part of the license to grow.