The open-weight AI model lever, not the chip race

The US-China AI race is being repriced on a different axis than the one Washington has spent two years trying to control. Capability is being shipped through open-weight releases, meaning trained model parameters anyone can download and run on their own hardware, while capital chases the proof that this lever can clear export curbs. The funding rounds landing in Beijing right now are the market's price on that question, not on chip access.

Bloomberg reports Moonshot, the Beijing-based lab behind the open-weight Kimi K3, hit US$300 million in annual recurring revenue by June, up from US$200 million in April. The release of K3's weights on July 27 triggered a US tech sell-off and another 'DeepSeek moment' round of repricing. Investors are now lining up a US$50 billion pre-money round before a Hong Kong IPO window that may open in 2026.

What appears to be happening — and what the ARR and valuation data suggest the market is pricing — is a shift in the binding constraint: when a Chinese lab can ship frontier-class capability without the chip access frontier labs rely on, the limiting factor becomes willingness to release rather than silicon availability. Moonshot executed that shift, the market responded, and capital is moving to lock in that proof before the next US export-curb round lands. The unresolved piece: the White House allegations of distillation and restricted-chip training, which Moonshot has not addressed. If those hold, the open-weight path closes. If they do not, this is the template the next round will copy.