Compute Is the New CitadelCompute Is the New Citadel
Machiavelli devoted an entire chapter of The Prince to fortresses — whether a prince should build them, and concluded the only true fortress is "not to be hated by the people," but that against foreign powers, physical chokepoints still matter. Translate this forward. The fortress of the AI era is the frontier training cluster, and the choke point is the lithography supply chain that feeds it.
When the Biden administration tightened export controls on advanced GPUs and EUV machines bound for China, the policy class debated it as trade economics — gains from trade, deadweight loss, the usual. Machiavelli would have laughed at the category error. Export controls are not economic policy. They are the modern equivalent of denying a rival city the iron for its arms and the engineers for its walls. He understood this intimately: as Secretary of the Second Chancery he spent years on the Pisa campaign, and his core lesson was that you do not win by hiring mercenaries (condottieri) whose loyalty is rented — you win by controlling your own arms. A nation that lets its strategic compute capacity be set by a foreign foundry it cannot defend has hired mercenaries. Taiwan is, in the most literal Machiavellian sense, an undefended armory holding the only weapons that matter. He would not be writing op-eds about TSMC diversification. He would be asking whether the armory can be held, and what happens the day it cannot.
At Bridgewater we drilled one discipline relentlessly: separate the narrative a market tells about itself from the mechanical cause-and-effect actually moving it. The AI policy narrative is "safety versus innovation." The machine underneath is "who holds the compute, who holds the weights, and is that entity loyal to American power." Machiavelli is the only analyst in the room who refuses to confuse the two.
Machiavelli devoted an entire chapter of The Prince to fortresses — whether a prince should build them, and concluded the only true fortress is "not to be hated by the people," but that against foreign powers, physical chokepoints still matter. Translate this forward. The fortress of the AI era is the frontier training cluster, and the choke point is the lithography supply chain that feeds it.
When the Biden administration tightened export controls on advanced GPUs and EUV machines bound for China, the policy class debated it as trade economics — gains from trade, deadweight loss, the usual. Machiavelli would have laughed at the category error. Export controls are not economic policy. They are the modern equivalent of denying a rival city the iron for its arms and the engineers for its walls. He understood this intimately: as Secretary of the Second Chancery he spent years on the Pisa campaign, and his core lesson was that you do not win by hiring mercenaries (condottieri) whose loyalty is rented — you win by controlling your own arms. A nation that lets its strategic compute capacity be set by a foreign foundry it cannot defend has hired mercenaries. Taiwan is, in the most literal Machiavellian sense, an undefended armory holding the only weapons that matter. He would not be writing op-eds about TSMC diversification. He would be asking whether the armory can be held, and what happens the day it cannot.
At Bridgewater we drilled one discipline relentlessly: separate the narrative a market tells about itself from the mechanical cause-and-effect actually moving it. The AI policy narrative is "safety versus innovation." The machine underneath is "who holds the compute, who holds the weights, and is that entity loyal to American power." Machiavelli is the only analyst in the room who refuses to confuse the two.