27 September 2026

Here is China’s main weakness on AI. Exploit it.

Washington Post | Thomas J. Duesterberg

President Donald Trump and Chinese leader Xi Jinping’s summit presents an opportunity to derail Beijing’s artificial intelligence strategy by exploiting its weak financial pillars. China’s aggressive attempt to capture global AI markets through state subsidies and open-weight models undercuts American investment returns, yet Chinese firms remain largely unprofitable and dependent on government support.

Domestic crackdowns on entrepreneurs and foreign investors have further dampened private funding, forcing flagship firms like DeepSeek to struggle for foreign capital and computing power. Meanwhile, Alibaba’s open-weight Qwen models achieved over 3 billion downloads in 2026, outpacing Meta’s Llama. The White House should ban Chinese models from U.S. markets until reciprocal access is established and renew limits on U.S. financing of AI infrastructure IPOs. If Trump deploys these economic tools, the United States can undermine Beijing’s advance.

Comment

Beijing's reliance on government-directed capital injections for firms like Z.ai masks an underlying inability to generate sustainable organic revenues without state-owned enterprise mandates. When flagship entities such as DeepSeek face private fundraising suspensions following leaked briefings on hardware deficits, it exposes the structural friction between state-directed industrial policy and the high-frequency capital requirements of advanced compute scaling. Consequently, venture capital flight resulting from regulatory crackdowns on domestic tech figures constrains the scaling velocity needed to close the hardware gap with American frontier labs.

Strategic Question for Discussion
What happens to the scalability of China's open-weight AI ecosystem if state-owned enterprise revenue streams can no longer offset the deepening operating losses of firms like Z.ai?
The pattern suggests that a contraction in state subsidies would force sudden consolidation among domestic AI developers. Without alternative venture capital inflows, these firms would likely curtail compute-intensive model training.
Share your assessment in the comments below.