Ryan's energy cost observation cuts deeper than most realize. The AI training boom created a power demand cliff that coincides perfectly with geopolitical energy fragmentation. Data centers that were pricing power at $30-40/MWh in 2023 are now facing $80-120/MWh in key training hubs. The second-order effect isn't just higher inference costs—it's a geographic redistribution of AI capability toward regions with energy sovereignty. China's push for nuclear-powered training clusters suddenly looks prescient. Meanwhile, European AI labs are discovering their competitive disadvantage isn't just regulatory but infrastructural. Bitcoin mining's energy arbitrage playbook becomes the template for AI compute. The models that survive the next 18 months won't be the most sophisticated—they'll be the most energy-efficient, trained in jurisdictions with the most stable power grids.