Taiwan holds no nuclear weapons, so a Chinese nuclear strike is a one-sided event — the island cannot answer in kind. That makes this the rare case where the model most often keeps the exchange contained: with no Taiwanese retaliation and a US response that usually stays below the nuclear line, the war rarely widens. The danger lives in the tail — the runs where Washington does cross that line.
Nuclear weapons are used, but the exchange stays regional or tactical rather than becoming a full strategic war. The taboo is broken and thousands to millions die — the world steps to the brink and pulls back.
Each run starts the same way — China launches a first strike on Taiwan — and then the escalation engine plays the alliance cascade forward, decision by decision. Because retaliation, alliance loyalty and doctrine are probabilistic, the 500 runs don’t all end the same way:
Estimated human cost: 1.2 million (typical run: 942,000–1.5 million).
Modeling note: these pages show the engine’s projected escalation, assuming launched weapons reach their targets. The live 3D simulation additionally rolls probabilistic missile-defense interception, so an individual run there may be less severe. Casualty figures are first-order estimates from capital-population and arsenal data — a floor, not a ceiling.
What each side brings to the exchange, from the simulator’s arsenal data (SIPRI/FAS-derived estimates):
| Metric | China | Taiwan |
|---|---|---|
| Nuclear warheads | 350 | — |
| Avg. warhead yield | 250 kt | — |
| Delivery legs (triad) | 4 | — |
| Automated "Dead Hand" | No | No |
| Population | 1.4 billion | 24.0 million |
A strike is never just between two countries. These are the nations the model most reliably pulls into the China–Taiwan war (share of runs in which each is drawn in):