The AI Trade Is Underpricing China Risk
Adam Taggart | Thoughtful Money® Aug 21, 2026
Summary
The AI trade may be underpricing China's ability to deliver capable models at a fraction of Western compute costs.
Western AI capex is hitting compute, chip, and power bottlenecks that Chinese rivals don't face in the same way.
If LLMs converge to similar capabilities, AI competition will pivot from raw intelligence to price.
Companies adopting cheaper foreign AI must weigh who gains long-term access to proprietary corporate data.
A huge share of US market cap rests on AI capex earning attractive returns, making sector disappointment systemic.
Peter Tchir's takeaways
- Top takeaway Argues the AI trade is underpricing China risk, drawing a parallel to 2005 when cheap imports entered through the back door and hollowed out US industry.
- Says the promised AI productivity 'holy grail' has not panned out — compute costs are climbing because of data center and chip bottlenecks, and companies still need staff to fact-check AI output.
- Warns LLMs may converge to similar capability levels and that distillation/IP theft is undermining the US AI moat.
- Highlights structural US bottlenecks — electricity constraints and NIMBY-style rules — as advantages China does not share, and notes that China could mobilize a much larger engineering workforce if given marching-orders direction.
- Wants to underweight the AI sector while overweighting other areas, citing how dependent current US equity valuations are on AI-related capex delivering returns.