This image, based on recent EPOC research, shows the global distribution of AI compute from 2019-2025. The US overwhelmingly dominates, accounting for roughly 75% of aggregate performance. China now holds a distant second place, its share fluctuating and declining after GPU export controls tightened. This concentration is driven by a profound shift: AI compute is now overwhelmingly private, with industry controlling 80% of performance, up from 40% in 2019. While performance doubles every nine months, the report highlights unsustainable resource growth, with hardware costs and power needs doubling annually. Researchers predict that if trends continue, a leading AI system in 2030 could cost $200 billion and require an immense 9 gigawatts of power, equivalent to nine nuclear reactors. They suggest securing such power is the primary constraint, likely forcing a shift towards training models across multiple, distributed sites rather than single colossal clusters. Can the US maintain its lead as its supply chains begin to suffer unprecedented stress?