While Americans argue over who should pay for AI's electricity, a very different answer is taking shape across Asia and the Gulf: treat the whole thing as a planning problem for the state to engineer, not a market to let run wild. This week produced two clear examples of that instinct, one from China and one from South Korea, and together they show both the appeal and the limits of trying to build your way out of AI's power crunch from the top down.
China's framing, from National Energy Administration head Wang Hongzhi, is almost a slogan: "powering computing with electricity and promoting electricity with computing." In practice that means tethering new computing hubs in the sun-and-wind-rich western provinces directly to renewable bases, nesting data centers alongside microgrids in the crowded east, and nudging slow, delay-tolerant model-training runs into off-peak hours so the centers act as flexible shock absorbers for the grid. It helps that China has the raw supply to plan with. Its installed generating capacity has passed 4.01 billion kilowatts, more than the United States, European Union, India, Japan and Russia combined, and it added the most recent billion in just two years. AI, in turn, is being wired back into the grid to manage it, with models that forecast renewable swings and reroute power around faults.
South Korea's version has a name: the "electrostate." In a policy briefing to the president on Tuesday, the environment ministry laid out a plan to power its AI and semiconductor ambitions entirely with clean energy, bringing 100 gigawatts of renewables online ahead of schedule, expanding offshore wind, holding nuclear and small modular reactors as baseload, and, crucially, building transmission ahead of demand so new data centers can plug in immediately. It even folds in water: wastewater reuse, groundwater storage, mobile desalination. The minister's pitch was to become "the first country to realize a future where clean energy powers AI."
The catch is that generation is rarely the real bottleneck, and plans are easier to announce than to string across a country. India makes the point bluntly. Executives at a Mumbai summit this week argued the country already produces more power than it needs outside peak months; the constraint is last-mile transmission and distribution, plus the awkward fact that 80 percent of $50 billion in planned data centers are crowding into just three city corridors. Renewables, one executive noted, lack the mechanical inertia that keeps a grid stable, and there is still no requirement for operators to disclose how much power or water AI actually consumes.
The stakes are sharpest where the grid is thinnest. The International Energy Agency expects data-center electricity use to climb from 415 terawatt-hours in 2024 to 945 by 2030, and much of the new building is happening in the Global South. A single proposed one-gigawatt campus in Kenya was stalled because it would nearly match a national grid that supplies just three gigawatts. With 666 million people worldwide still lacking electricity at all, the paradox is stark: the same buildout that could help close development gaps could just as easily deepen them. Top-down planning may beat the American scramble, but only if the blueprints survive contact with the wires.