Article · 2026-07-20 · 6 moments

Who’s Afraid of Chinese Models?

Everyone is worried about Chinese models, but the frontier labs will be fine; we need to enable open U.S. alternatives. ✦ AI generated

Read original ↗

GUEST Ben Thompson

Claim◆

As intelligence becomes commoditized, Anthropic and OpenAI will be fine because they already have among the lowest costs per unit of frontier-quality intelligence, and being on the frontier positions a lab to dominate non-frontier markets as well.

Thompson's central economic argument: frontier labs' superior cost structure and head-start on capability mean they should thrive even as intelligence commoditizes, making fears about Chinese models overblown.

transcript

Ben Thompson: Anthropic and OpenAI likely have among the lowest costs per unit of frontier-quality intelligence, thanks to model capability, serving scale, and token efficiency. They are serving models at a particular capability level for months before their competitors, and are simultaneously applying the best models to optimizing those costs.

rebuts · 1

Claim

Because U.S. open weight model makers must obey frontier labs' terms of service barring distillation, they end up worse than Chinese alternatives and distill Chinese models instead — the U.S. should legalize training-data collection as fair use and bar anti-distillation terms of service.

Thompson argues distillation gives Chinese labs a recurring structural edge over Western open-weight makers bound by frontier labs' anti-distillation terms, and calls for U.S. policy change to fix this.

transcript

Ben Thompson: This is a point that bears repeating: because U.S. open weight model makers must follow the frontier labs' terms of service, they (1) are worse than Chinese alternatives and (2) end up distilling the distillation, just with a detour through Chinese labs. Wouldn't it be better if western open weight model makers could go to the source?

provides context · 1

Context

Marginal costs are back in a big way for AI, both in the short-term implications of state-of-the-art free/cheap models and in the long-term structure of the industry.

Thompson frames the debate over Kimi K3 as evidence that old cost-structure principles (once thought irrelevant to zero-marginal-cost software) are returning to the center of AI industry dynamics.

transcript

Ben Thompson: That was never more apparent than this past weekend, when arguments raged on X about the implications of Kimi K3, another open weights model out of China, approaching the state-of-the-art in terms of capabilities. The long and short of it is this: marginal costs are back in a big way, both in terms of short-term implications of state-of-the-art free models, and in terms of the long-term structure of the industry.

extends · 1supports · 3

Mechanism

Open weight models are not truly free to use — the 'free' only refers to R&D, a fixed sunk cost, while COGS from running inference is real, revenue-correlated, and applies just as much to Kimi as to any other model.

Thompson distinguishes R&D (a fixed cost avoided by using open weights) from COGS (variable inference cost that scales with revenue and is unavoidable for any model, open or closed).

transcript

Ben Thompson: What is related to revenue is COGS — cost of goods sold — and COGS is real for AI in a way it hasn't been for software for a very long time. Specifically, running inference on a model — whether that model be Kimi or Fable — costs money, and the amount of money an AI provider spends on inference is, at least in most business models, directly correlated to revenue.

provides context · 1rebuts · 1

Claim◆

The real reason to be afraid of Chinese models is cybersecurity: U.S. restrictions bar defenders from using Fable/Sol for cybersecurity, pushing them toward Chinese models instead — the U.S. should loosen those restrictions and put U.S. open weight makers on equal footing with China.

Citing Hugging Face's use of a Chinese open model to respond to a breach because U.S. frontier models were restricted for cybersecurity use, Thompson argues this policy is dangerously self-defeating and U.S. restrictions should be loosened.

transcript

Ben Thompson: Right now defenders are effectively banned from using Fable or Sol for cybersecurity because of Trump administration directives; that means the best alternative is using models from a country which has been trying to weaken our cyber defenses for years. This is insane! The better course is clear: first, loosen Fable and Sol restrictions on cybersecurity, and second, ensure that U.S. open weight model makers are on an equal playing field with China.

Mechanism

Tokens are not a commodity, since different models need different amounts of tokens to reach a correct answer; what is actually fungible is the intelligence (the correct answer) that tokens produce, not the tokens themselves.

Thompson argues against Nvidia's 'token factory' framing, contending intelligence (correct outputs), not raw tokens, is the true fungible commodity, since models vary widely in tokens needed per answer.

transcript

Ben Thompson: A token from one model, however, is not the same as a token from another model. What is fungible is what is constructed from tokens, which is to say intelligence. In other words, if both Kimi and Sol generated the right answer, then that answer is fungible; the difference in tokens generated to get to that right answer is a contributor to a difference in COGS.

extends · 1gives example · 2supports · 2

Highlight slides

The Real China-Model Risk: Cybersecurity✦ from: The real reason to be afraid of Chinese models is cybersecurity: U.S. restrictions bar defenders from using Fable/Sol for cybersecurity, pushing them toward Chinese models instead — the U.S. should loosen those restrictions and put U.S. open weight makers on equal footing with China.Frontier Labs' Cost Advantage✦ from: As intelligence becomes commoditized, Anthropic and OpenAI will be fine because they already have among the lowest costs per unit of frontier-quality intelligence, and being on the frontier positions a lab to dominate non-frontier markets as well.Thompson's Fix✦ from: The real reason to be afraid of Chinese models is cybersecurity: U.S. restrictions bar defenders from using Fable/Sol for cybersecurity, pushing them toward Chinese models instead — the U.S. should loosen those restrictions and put U.S. open weight makers on equal footing with China.Why Commoditization Won't Hurt Them✦ from: As intelligence becomes commoditized, Anthropic and OpenAI will be fine because they already have among the lowest costs per unit of frontier-quality intelligence, and being on the frontier positions a lab to dominate non-frontier markets as well.

Related episodes

Video · The Cognitive Revolution

Fable's Back, AI Engineer Recap, & SambaNova

→

Video · a16z Podcast

Ben Horowitz on the Global Race for Tech, Power, and Influence

→

Video · All-In Podcast

OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

→