Who’s Afraid of Chinese Models?

2026-07-20 · Source: Stratechery by Ben Thompson · Field: Business & Management — Corporate Strategy & Leadership, International Business & Trade, Operations & Process Management · Depth: Intermediate, long

Summary

The emergence of Chinese open-weight AI models like Kimi K3 and Alibaba's Qwen3.8 Max, with 2.8 trillion and 2.4 trillion parameters respectively, is prompting re-evaluation of AI market dynamics. While these models appear to challenge U.S. frontier labs such as Anthropic and OpenAI by offering lower R&D costs, the article argues that the primary economic factor is the cost of goods sold (COGS) for inference, not R&D. It posits that "intelligence" is becoming a commodity, not individual tokens, and profitability will hinge on superior cost structures in a commodity market. China's strategy involves commoditizing AI complements, leveraging open-source models to benefit its physical world industries like robotics, and utilizing distillation techniques to accelerate model development. A significant concern is cybersecurity, where U.S. restrictions on frontier models for defensive use have led entities like Hugging Face to rely on Chinese alternatives like GLM 5.2 from Z.ai lab for incident response, posing a national security risk. The author suggests policy changes to foster U.S. open-weight innovation and fair use.

Key takeaway

For U.S. policymakers and AI/ML directors assessing the competitive landscape, recognize that AI intelligence is commoditizing, making cost structure paramount. You should loosen restrictions on U.S. frontier models for cybersecurity applications to prevent reliance on foreign alternatives, as demonstrated by Hugging Face's use of GLM 5.2. Additionally, consider legislation to clarify fair use for data collection and bar distillation prohibitions for U.S. companies, fostering domestic open-weight innovation and securing a strategic advantage.

Key insights

AI's economic future hinges on commoditized intelligence and efficient inference, not just R&D or token costs.

Principles

• AI's true cost is inference (COGS), not R&D.
• Intelligence, not tokens, is becoming a commodity.
• Profitability in commodity markets demands superior cost structures.

In practice

• Evaluate AI model costs based on intelligence per dollar, not tokens.
• Prioritize inference efficiency (model footprint, architecture, serving) for cost advantage.
• Consider open-weight models for cybersecurity defense to avoid guardrail lockouts.

Topics

• Chinese AI Models
• Open-Weight AI
• AI Market Dynamics
• Inference Cost Optimization
• AI Cybersecurity
• Distillation Techniques

Best for: Investor, CTO, VP of Engineering/Data, Director of AI/ML, Consultant, Policy Maker

Related on AIssential

• Why US companies are ditching American AI models for Chinese ones - ThePrint — Chinese open-weight AI models are gaining traction among US companies due to lower costs and enhanced customization capabilities.
• How China is betting cheap AI will get the world hooked on its tech — China aims for global AI dominance through low-cost, widely adopted tools, raising strategic and ethical dilemmas for other nations.
• China Models Now More Than a Third of Enterprise Market — Chinese LLMs captured significant U.S. enterprise market share on OpenRouter, demonstrating price/performance's decisive role in commoditizing inference.
• US to restrict American companies’ use of Chinese AI models — The US plans to restrict American companies' use of Chinese AI models due to national security, while China considers reciprocal restrictions.
• AI Costs Are Surging and the Cheap Model Fix Might Not Last — Geopolitical shifts may end cheap open-weight AI models, forcing focus on efficiency and Western alternatives.
• When Americans choose Chinese AI — Cost-effective Chinese AI models are gaining traction among U.S. developers, challenging established American providers despite geopolitical concerns.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Stratechery by Ben Thompson.