Key Takeaways
- Chinese research labs now build roughly 95% of the highest-quality open-weight models used by developers worldwide.
- China is pairing open software dominance with a vertically integrated domestic hardware supply chain, including chips from Huawei.
- Private venture capital in the United States cannot match state-directed hardware subsidies on its own; competing requires federal industrial policy.
- Relying on foreign open-source weights creates a hidden architectural lock-in for American startups that build their products on top of them.
The Open-Source Software Monopoly
Silicon Valley likes to believe it owns developer mindshare. Sean Lie sees a different reality.
“The open source model market is 100% Chinese, right? Almost. Okay. 95%,” Lie observed. “Most of the big open models that are high quality are coming from the Chinese labs.”
When American AI labs pivoted toward closed APIs, paywalls, and safety filters, Chinese organizations filled the vacuum. They flooded public repositories with high-performing, permissively licensed weights. Western founders downloaded them by the millions. They fine-tuned them, ran evaluations, and built commercial applications on top of them.
Lie notes the double-edged reality: “In some ways, it's great that sharing is happening and the global community is benefiting. But also obviously, that's a very strategically challenging place to be, to have such dependence on them on the model.”
If your core engine depends entirely on model weights designed, trained, and filtered abroad, you do not control your product roadmap. You inherit their architectural choices, their data biases, and their security vulnerabilities.
The Silicon Trap Behind the Weights
Software is only half the offensive. Open-weight models are not created in an abstract cloud; they require massive compute clusters to train and deploy. Lie pointed out that China is building domestic silicon specifically tuned to run these workloads.
By developing custom silicon through companies like Huawei alongside domestic fabrication capabilities, China is insulating its tech sector from Western export controls. When the base models and the underlying silicon evolve together, the software optimizes for that exact silicon. If Western startups rely purely on foreign weights while Western chipmakers focus strictly on proprietary platforms, American builders risk falling behind on efficiency.
“These guys know what they're doing,” Lie said. “It's something that I believe is not solved by one company or one fab. It needs government level, national interest level kind of initiative to be able to make this happen.”
The market cannot fix this balance through standard venture funding rounds. When foreign rivals operate with state backing across both silicon fabs and open models, private companies need deliberate, coordinated policy support to compete at scale.
What to Do With This
Audit your dependency on foreign open-weight architectures before the end of the quarter. If your core product relies on a model trained by an overseas lab, build an automated evaluation pipeline to test your workflows against domestic open alternatives or internal fine-tunes on standard US hardware stacks.