World War AI (www.epsilontheory.com)

The piece argues that the current AI datacenter and infrastructure buildout is less a gentle productivity revolution and more a full-scale economic mobilization — a "World War AI" — that will require war‑like reallocation of capital, labor and energy. Citing narrative tracking (Perscient’s AI-capex/China narrative z‑score hit a record 3.96) and JPMorgan forecasts, the author says the U.S. is on track to spend trillions over the next four years (roughly the inflation‑adjusted $4–5 trillion the U.S. spent on WWII, or about 15% of today’s GDP) on AI/datacenter expansion. Corporate requests for a federal backstop on datacenter debt (notably OpenAI CFO Sarah Friar’s comments) and a $1.4T line item for “alternative capital/governments” in JP Morgan’s model signal that large public subsidies, guarantees or direct spending are likely. Technically and economically, this implies massive crowding out: private and public capital flowing into hyperscalers will raise borrowing costs and choke credit for consumers and SMEs. Early signals already show consumer credit rejections at decade highs and mortgage refi denials north of 45%. Through H1, tech spending accounted for nearly half of U.S. GDP growth while datacenter electricity demand is projected to surge (JPMorgan ~1,100 TWh globally), stressing energy systems. The practical implication for the AI/ML community: accelerated capacity and funding for large models, but an ecosystem that could deepen inequality, concentrate compute and talent, and depend heavily on state guarantees and energy expansion to sustain exponential compute demand.