The Trojan horse

AI is not going to win a war with China by itself. Treating it as a substitute for ships, magazines, yards, energy, and magnets is the Trojan-horse part. China is running both tracks. The United States is running the software track hard and the metal track late.

The question is not whether models matter. They do. The question is whether processing power is the scarce strategic resource. For the war people actually game in the Western Pacific, the scarce resources are magazines, yards, magnets, transformers, and time.

What AI actually buys

In a Taiwan or Western Pacific fight, AI is a multiplier, not a magazine. It compresses targeting, ISR fusion, logistics, planning, and swarm coordination. Both sides are already doing this. The United States has used commercial models in recent operations. China is fielding AI across command, sensing, drone swarms, and counter-U.S. awareness at sea and in space.

The frontier-model gap is measured in months, not years. China closes it with a fraction of U.S. private spend by distilling American models and shipping cheap open-weight systems. That matters. It does not sink a destroyer. It does not refill a vertical-launch cell. It does not replace a yard that can turn hulls in months instead of years.

There is no durable software moat. Capabilities leak through distillation, cloud access, smuggling, and open weights. An advantage measured in quarters is not a war-winning monopoly. A nuclear monopoly lasted years. This one will not.

What finite resources are actually producing

Ships. China holds an order-of-magnitude shipbuilding advantage. Civil and military yards are fused. PLAN hull count already exceeds the U.S. battle force. The U.S. Navy sits near 290 ships.

Missiles and mass. U.S. stocks of the munitions that would be fired in the first weeks of a Pacific war were already thin and got thinner after recent operations. Production is being surged. Lead times are still years.

Energy. China adds generation at a national scale. The United States is bolting gigawatt campuses onto a grid that was not designed for them.

Materials. Rare earths and magnets are a live licensing weapon. You cannot train your way out of a magnet shortage.

Every megawatt locked into a hyperscale training cluster is a megawatt that is not firming a munitions plant, a yard, or a base that has to fight in the dark.

Four traps

Opportunity cost dressed up as strategy. Funding the opponent’s catch-up through distillation and cloud access. Concentration and dependency on Taiwanese logic and Chinese batteries, grid gear, and magnets. Narrative capture: “win the AI race” is a fundraising line, not a force-planning construct.

Bottom line

As a theory of victory, the AI-first bet is a Trojan horse. As a tool, AI is real. Put models on systems you can actually build and replace. Do not treat the next model drop as a substitute for the next dry dock.

The war is not won by the cluster with the most accelerators. It is won by the side that still has missiles and ships after the first exchange, and enough electricity and metal to make more.