Accused Of Copying U.S. AI, Chinese Founders Mint Billions

Key Concepts

• Distillation: A machine learning technique where a smaller, more efficient model is trained to mimic the outputs and behaviors of a larger, more powerful "frontier" model.
• Frontier Models: Large-scale, highly capable AI models (e.g., Claude, GPT-4) that represent the current state-of-the-art in performance.
• Open-Source AI: Models whose architecture and weights are publicly available, allowing for rapid adoption and modification.
• Meme Stock Revenue Multiples: Extremely high valuation ratios (e.g., 400x–550x revenue) that reflect speculative investor enthusiasm rather than traditional fundamental value.
• Token Economics: The cost structure of AI models, measured by the price per input and output "token" (units of text processed by the model).

1. Allegations of Intellectual Property Theft

Major US AI labs, specifically OpenAI and Anthropic, have formally accused Chinese AI companies—including DeepSeek, MiniMax, and Moonshot AI—of systematically siphoning proprietary capabilities.

• Methodology: The accused companies allegedly used approximately 24,000 fraudulent accounts to prompt Anthropic’s Claude models 16 million times.
• Objective: To extract "distilled" data (coding, reasoning, and behavioral patterns) to train their own competing models.
• Evidence: Google’s Threat Intelligence Group has corroborated these concerns, issuing a report warning of an increase in "distillation attacks" targeting models like Gemini.

2. The Economics of Distillation

Distillation is the core technical driver behind the current competitive shift.

• Efficiency: It allows smaller models to achieve 80% to 90% of the performance of a frontier model while requiring significantly less computing power.
• Cost Advantage: Chinese models are significantly cheaper. For example, Zhipu AI’s GLM-5 model reportedly costs five times less per input token and ten times less per output token than Anthropic’s Claude Opus.
• Market Impact: This price disparity creates a "kill line" for expensive, closed-source models. As Jenny Xiao (Lightspeed Capital) notes, "You don’t need the smartest model to do average enterprise workloads," suggesting that if performance gaps narrow, customers will prioritize cost over marginal intelligence gains.

3. Financial Landscape and Market Valuation

Despite the controversy, Chinese AI founders are seeing massive wealth accumulation and investor interest.

• Wealth Creation: Several founders have reached billionaire status, including Wang Junjie (MiniMax, $7.1B), Lu (Zhipu AI, $8.7B), and Liang Wangfeng ($11.5B).
• Market Volatility: Chinese AI firms are trading at extreme revenue multiples (400x–550x), which are more than double the valuations assigned to US counterparts like OpenAI and Anthropic.
• US vs. China: While US labs hold private valuations in the hundreds of billions ($380B for Anthropic, $840B for OpenAI), these remain untested by public markets. Conversely, the recent IPOs of companies like MiniMax and Zhipu AI in Hong Kong have introduced significant volatility and price pullbacks.

4. Key Perspectives and Quotes

• John Holdquist (Google Threat Intelligence): "It’s not easy to build these models, and distillation is a way to leapfrog that process."
• Jenny Xiao (Lightspeed Capital): "Open-source models are essentially a kill line."
• Martin Casado (Andreessen Horowitz): Noted that approximately 80% of American startups currently using open-source models are relying on Chinese-developed ones, highlighting the low friction and minimal switching costs for developers.

5. Synthesis and Conclusion

The AI industry is currently facing a fundamental tension between the high-cost, closed-source development model favored by US labs and the rapid, cost-efficient, and often open-source approach adopted by Chinese firms. Distillation has effectively flattened the competitive landscape, allowing smaller players to bypass the massive R&D costs of building frontier models from scratch.

The long-term viability of the AI sector remains uncertain. While US labs argue that their intellectual property is being stolen, the market is increasingly favoring cheaper, accessible alternatives. The ultimate test for the industry will be whether closed-source companies can prove a sustainable economic advantage before the "meme stock" valuations of their competitors face a reality check or before open-source performance renders their proprietary models obsolete for standard enterprise use.