Capital Flows
· Capital Flows
· July 03, 2026 at 02:59
· ⏱ 2 min read
| Read on Substack ↗
Summary
The article argues that China's strategy of subsidizing cheap AI models (10-100x cheaper) mirrors Amazon's playbook and threatens the entire AI trade, which is the most crowded trade in history funded on borrowed money. The market is vulnerable to a repricing event like DeepSeek's January 2025 crash that erased $589 billion from a single stock in a day, because export controls cannot stop open-source model proliferation and the US has a power grid disadvantage.
•China runs AI models one generation behind in quality but priced 10 to 100 times cheaper, subsidized by the state so they never need the API margins US labs depend on.
•When the US suspended its strongest model on June 12, a Chinese lab open-sourced a near-frontier rival the next day, illustrating the futility of export controls on model weights.
•The US has chips but a tight power grid, while China has an eight times electricity advantage but only about 4% of NVIDIA’s compute, turning AI into an energy business.
•DeepSeek in January 2025 erased $589 billion from one stock (NVIDIA) in a single day, and the same loop can repeat with far more premium priced in today.
The article directly references the DeepSeek event that wiped $589 billion from NVIDIA's market cap in a day and warns that the same loop can run again with more premium priced in, signaling vulnerabi
The article directly references the DeepSeek event that wiped $589 billion from NVIDIA's market cap in a day and warns that the same loop can run again with more premium priced in, signaling vulnerability to price disruption from cheaper Chinese AI models.
Risk: A repeat of the January 2025 selloff could be amplified by the current crowded positioning and leverage in the AI trade.