Every AI Constitution Has an Energy Clause
The politics of AI will be written not only in model rules, but in who supplies the electricity, water and grid capacity.
Artificial intelligence is discussed as if it lives in an abstract layer above ordinary politics. It does not. Every query reaches physical processors. Every processor draws electricity and releases heat. Every large facility occupies land, connects to transmission and chooses a cooling system.
The next AI constitution therefore has an energy clause, whether governments write one or allow private contracts to write it for them.
A global demand with local consequences
The International Energy Agency estimated that data centres used about 415 terawatt-hours of electricity in 2024, around 1.5 percent of global consumption, and projected roughly 945 TWh by 2030. Its 2026 update says AI-focused facilities increased electricity use by about 50 percent in 2025.
Those figures do not justify claims that AI will consume the world's grid. Motors, cooling, vehicles and industrial growth remain larger sources of new demand in many economies. AI may also improve forecasting, maintenance and materials discovery.
Global shares, however, hide local concentration. A cluster of campuses can require as much firm power as a city and can arrive faster than new transmission. The IEA has warned that roughly a fifth of planned data-centre projects could face delays if grid risks are not addressed.
The political conflict emerges at the connection point. Who pays for the new substation? Does an industrial customer receive priority over housing or manufacturing? Is a renewable contract adding clean supply or relabelling electricity that would have served somebody else? Which community carries water stress and backup-generator pollution?
These are distributional questions. Leaving them to confidential negotiations does not make them less political.
Efficiency does not end the argument
The energy required for a simple AI task has been falling rapidly. That should be welcomed. Yet cheaper computation encourages more use, and new reasoning, video and agentic tasks may require hundreds or thousands of times the electricity of a short text response.
Efficiency is therefore a variable, not an exemption. A responsible system measures both energy per task and total demand. It distinguishes a useful small model from a prestige training run, and a flexible workload from a service that must operate continuously.
Governments should resist false precision. Estimates differ because model design, chip utilisation, cooling, regional grids and user behaviour change quickly. Policy should work across a range of demand rather than depend on one dramatic forecast.
Additional demand should bring additional capacity
A simple principle can organise the bargain: major AI facilities should finance credible additional power, grid upgrades and flexibility in proportion to the new demand they create.
Additionality need not prescribe one national energy doctrine. Some regions can combine wind, solar and storage. Others may rely on nuclear, hydro or other firm sources. Workloads such as training can sometimes move to periods or locations with abundant power. The obligation is to make the incremental burden visible and avoid quietly transferring it to ordinary ratepayers.
Regulators should publish connection queues, upgrade costs and delivery dates. Facilities receiving public support should disclose comparable power- and water-use measures, their source mix and the share of demand that can respond to grid conditions. Subsidies should reward low-water cooling, heat reuse and verifiable clean additions rather than attractive certificates.
The strongest countercase
Special rules for AI can turn into symbolic punishment. Data centres carry hospitals, banking, communications and ordinary cloud services as well as model workloads. A rigid classification will be gamed, and onerous disclosure can drive investment to jurisdictions with dirtier power and weaker safeguards.
The answer is to regulate load characteristics, not fashion. A large, concentrated, continuous electricity demand should face consistent connection and disclosure rules whether it serves AI, cryptocurrency, streaming or another digital service. AI deserves attention because it is accelerating the scale and speed, not because computation is morally suspect.
There is also a development case. Regions with abundant clean energy may use data centres as anchor customers that finance transmission and storage. That bargain can be good—if the infrastructure remains useful beyond one campus and the public knows who paid and who benefits.
Infrastructure is governance
AI policy is often framed around safety evaluations, copyright and competition. All matter. But a rights-respecting model running on a secretive infrastructure bargain is not fully governed. Citizens experience the system through bills, land use, water access and reliability as well as through its outputs.
No country should have to choose between digital capability and energy justice. It can require builders to bring capacity, reveal impacts and share the infrastructure value they create.
The physical world is not a side constraint on artificial intelligence. It is the place where promises become costs. Every AI constitution has an energy clause. Democratic government should write it in public.
The Global Federation examines technology through the institutions beneath it: authority, public cost, accountability and the terms on which progress is shared.