For much of the past decade, the global race for artificial intelligence supremacy has been framed as a contest over algorithms, talent, and semiconductor chips. Governments and corporations alike have treated access to advanced processors as the decisive factor separating leaders from laggards. Export controls, industrial subsidies, and geopolitical maneuvering have all revolved around who designs, manufactures, or controls the most powerful chips. Yet this focus, while understandable, is increasingly incomplete. As the scale of AI systems explodes, a more basic input is emerging as the true strategic choke point: energy. In the coming years, reliable and affordable electricity will determine which countries can sustain large-scale AI development and deployment-and which will fall behind.

This shift is best understood through the lens of political economist Albert O. Hirschman, who argued in National Power and the Structure of Foreign Trade (1945) that national power derives from control over bottlenecks that others depend upon. In today’s AI ecosystem, the United States has exercised power by dominating chip design and restricting China’s access to cutting-edge semiconductors. China, for its part, has countered through leverage over rare-earth materials essential for chips, magnets, and advanced electronics. But as AI models grow ever larger and more computationally intensive, the bottleneck is moving downstream. Chips matter, but chips without electricity are inert. Data centers without abundant power are simply expensive warehouses.

The energy intensity of AI is staggering. Training large language models, running inference at scale, and supporting cloud-based services require vast and continuous flows of electricity. Data centers must operate 24/7, with minimal interruptions and tight tolerances for voltage fluctuations. According to the International Energy Agency, roughly 20 percent of planned global data-center capacity could be at risk by 2030 due to grid bottlenecks and delays in connecting new facilities to power networks. This is........

© Blitz