The entire AGI bet rests on a single island - and the market doesn't seem to care

I've been thinking about concentration risk in the AI trade and I keep coming back to the same bottleneck that nobody prices in. ([AGI Industrial Chain](https://www.bearbull.io/blog/Deep-Dive/AGI-Industrial-Chain))

Every major AI accelerator - NVIDIA H100s, Google TPUs, Amazon Trainium - is fabless. They design the chips but outsource manufacturing entirely. TSMC produces roughly 90% of the world's leading-edge semiconductors, and their advanced fabs are almost entirely in Taiwan, 180 km from mainland China.

The hyperscalers (Microsoft, Google, Meta, Amazon) have committed a combined $300B+ in AI CAPEX over the next few years. That spend assumes the supply chain stays intact. There is no viable near-term alternative: Intel Foundry is years behind on yield, Samsung's advanced node output is a fraction of TSMC's.

A Taiwan Strait escalation - or even a sustained blockade - wouldn't just hit TSM stock. It would freeze the hardware supply chain for every company in the AI space simultaneously. The CAPEX already deployed becomes stranded. Model training timelines slip. The entire investment thesis gets repriced.

I don't see this reflected in valuations anywhere. NVDA trades at 35x forward earnings. The hyperscalers are at all-time highs. TSM itself is up significantly despite the geopolitical backdrop.

Is this actually priced in somewhere I'm not looking - or is the market just ignoring the physical reality of how these models get built?

*Disclosure: Long TSM, NVDA*