While Western AI labs debate safety theater, China's foundation model companies are already trading on public markets.
The Summary
- Azeem Azhar spent a week meeting Chinese AI teams, including publicly listed foundation model companies Zhipu and MiniMax, plus Kimi, Alibaba, Xiaomi, and Bytedance
- Two Chinese LLM builders have already gone public while Western counterparts remain private and venture-backed
- The gap between Chinese AI deployment velocity and Western AI caution is widening into a structural advantage
The Signal
Zhipu and MiniMax are publicly traded foundation model companies. Not "planning to IPO." Not "exploring strategic options." Trading. While OpenAI burns through capital raises and Anthropic courts enterprise customers, China's model builders are answering to public shareholders.
This isn't about who builds the smartest models. It's about who builds sustainable businesses around them faster. Public markets force discipline. They punish vaporware. They reward revenue and margins. The fact that two Chinese foundation model companies cleared that bar while every major U.S. player is still private tells you something about commercial maturity.
"China's AI companies are optimizing for markets, not manifestos."
The Western narrative positions Chinese AI as derivative or censored. But derivative tech that ships at scale beats groundbreaking tech stuck in beta. And censorship concerns look quaint when the alternative is U.S. companies that won't release models for fear of misuse, won't train on certain data for fear of lawsuits, and won't deploy certain features for fear of regulators.
Azhar's tour included Kimi (the AI assistant with 30M+ users), Alibaba (whose Qwen models are open-source and widely deployed), Xiaomi (shipping AI into phones and cars), and Bytedance (which runs the world's most successful AI-driven content platform). These aren't research labs. They're product companies with distribution.
Key differences in approach:
- Chinese models train on everything available, legal questions dealt with later
- Deployment timelines measured in weeks, not quarters
- Safety is a feature constraint, not an existential threat narrative
The U.S. advantage in cutting-edge research is real but shrinking. The Chinese advantage in turning models into products people actually use is growing. That's the gap that matters for the agent economy. Agents need infrastructure, distribution, and commercial models that work. Public markets accelerate all three.
The Implication
Watch where the capital flows next. If Chinese AI companies can raise public money while Western labs burn venture dollars, the cost of capital alone becomes a competitive moat. Lower capital costs mean more experiments, faster iteration, cheaper inference.
For builders in the West: stop waiting for the perfect model or the perfect regulatory framework. The companies winning in China aren't smarter. They're faster. Speed is a feature when the technology is moving this quickly.