Elon Musk's xAI signs a gas turbine procurement deal with South Korea, revealing the structural power bottleneck faced by global AI data centers. The real constraint isn't factory capacity but rare elements—key materials like rhenium, tantalum, tungsten, and other nickel-based single-crystal alloys. Global annual production is only in the tens of tons, with an expansion cycle of 10-15 years. As AI computing power demand grows exponentially, 'dispatchable, quickly deployable baseload power' has become a scarce resource. Renewables like wind and solar are unstable, nuclear takes too long, and in the short term, natural gas + gas turbines are the only practical solution. The value of South Korea's Doosan lies not in scale but in structure—it’s not bound by the U.S. military-industrial supply chain, has a sufficient supply of engineers, and is better suited for the custom needs of AI data centers. AI is pushing the world from business cycle problems to physical limit problems.