Who’s Afraid of Chinese Models?
I did not, as I usually tell the story, find this very satisfactory: to me the nature of tech, particularly the fact that software and distribution had zero marginal costs (and zero transaction costs), was something fundamentally different; putting in zeroes in formulas tends to wreak havoc! I soon realized, however, that that was my opportunity. The fundamental insight undergirding Aggregation Theory is that zero marginal costs leads to fundamentally different value chains than people once expected from the Internet: centralization and scale in a world where controlling demand mattered more than distributing supply.
What is fascinating about AI, however, is the extent to which those old universal principles are coming back to the forefront. 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.
COGS Versus R&D
One of the most common misconceptions undergirding discussion of open weights models is that they are cheaper — free, even. After all, you can just download the weights, and skip the time and expense and capabilities necessary to create your own model. That is, of course, true, but the “free” in this case is a reference to the amount you need to spend on research and development; R&D is a fixed expense that is independent of the revenue you generate. If you spend $1 million in R&D, it doesn’t matter if you do $100 thousand in revenue or $100 million; you still spent $1 million on R&D (it does, of course, impact your profitability).
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. To reuse the above example, generating $100 million versus $100 thousand in revenue will likely require 1,000x COGS. In concrete terms, if it costs 50 cents to generate the tokens that drive $1 in revenue, then $100 million in revenue will have $50 million in COGS; $100 thousand in revenue will only have $50 thousand in COGS.
The point in terms of open weight models is that they are not free to serve. Kimi K3 costs $3 per million input tokens, and $15 per million output tokens; that is cheaper than Sol’s $5 per million input tokens and $30 per million output tokens, but that might not even be the right measurement.
Tokens Versus Intelligence
Nvidia CEO Jensen Huang has described what Nvidia is building as “token factories”, and from Nvidia’s perspective that framing makes sense. Nvidia GPUs are model agnostic: they generate tokens, and do so in the fastest and most efficient way possible. That leads to measurements like tokens-per-second, time-to-first-token, tokens-per-watt, token cost, etc., and Huang argues that these metrics will be the basis for decision-making.
Summary: Who’s Afraid of Chinese Models?
Ben Thompson argues that investors are overreacting to the rapid progress of Chinese artificial intelligence models like Kimi K3 and Alibaba’s Qwen3.8 Max. While these models appear much cheaper than leading American models from OpenAI and Anthropic, he believes the pricing advantage is misleading.
The key point is that the AI industry is gradually shifting from being a race to build the smartest model toward becoming a business of producing intelligence as efficiently as possible.
What are “tokens”?
For people unfamiliar with AI, a token is simply a small unit of language that an AI model processes. It isn’t exactly a word. Sometimes a token is a whole word (“market”), while other times a long word might be split into several tokens.
For example: “The stock market rallied today” might be broken into roughly six or seven tokens. Every question you ask an AI and every word it generates is measured in tokens.
Think of tokens as being similar to:
- gallons of gasoline for a car
- kilowatt-hours for electricity
- pages printed by a printer
The more tokens an AI must process to answer a question, the more computing power—and therefore money—it costs.
That is why AI companies charge by the number of input and output tokens rather than by the question itself.
Why token prices can be misleading
Many investors compare AI models simply by looking at the price per million tokens.
The author says this is the wrong comparison.
Imagine two mechanics:
- Mechanic A charges half as much per hour but takes four hours to fix your car.
- Mechanic B charges twice as much per hour but finishes in one hour.
Mechanic B is actually cheaper overall.
AI models work similarly.
A Chinese model may charge fewer dollars per million tokens, but if it needs twice as many tokens to reach the same answer, it may actually cost just as much—or even more—to operate. What really matters is the total cost of producing useful intelligence, not the price of each individual token.
Intelligence is becoming a commodity
The author believes AI is gradually becoming more like commodities such as oil or electricity.
When multiple companies can produce intelligence of similar quality, customers will simply buy from whichever supplier delivers it most cheaply.
That shifts the competitive advantage away from flashy model announcements and toward operational efficiency.
Companies that can produce intelligence at the lowest cost should earn the highest long-term profits.
Why American AI companies are still doing well
Despite Chinese competition, companies like OpenAI and Anthropic currently enjoy two major advantages:
- Their models remain the most capable.
- There is still far more demand than available computing capacity.
Because demand exceeds supply, they can charge premium prices today.
The author believes those high prices are largely the result of today’s compute shortage rather than permanently superior economics. As more AI chips and data centers come online, prices are likely to fall.
Why China is releasing open models
The article argues that China’s strategy is not simply to compete with OpenAI.
Instead, China wants to make advanced AI widely available through open-weight models that anyone can use.
Doing so encourages businesses around the world to build products using Chinese AI technology, making AI cheaper and accelerating industries where China already has major strengths, such as manufacturing and robotics.
The “distillation” controversy
The author also discusses AI distillation, where one AI model learns by studying the answers produced by another, more advanced model.
Critics say Chinese companies benefit from learning from American models.
The author acknowledges that this probably helps Chinese developers but argues that knowledge sharing has always driven technological progress. He even suggests the United States should change its laws to make this process easier for American companies instead of trying to block it.
The author’s biggest concern: cybersecurity
Ironically, the author says the greatest risk is not that Chinese AI will dominate commercial markets.
Instead, he worries that U.S. restrictions on using the most advanced American AI models for cybersecurity are forcing companies to rely on Chinese open-source models to defend against cyberattacks.
He argues that this is strategically backwards and that American defenders should have access to America’s best AI tools.
Bottom line for investors
The article suggests that investors should look beyond headlines claiming that Chinese AI models are “cheaper.”
The more important question is which companies can produce the most useful intelligence at the lowest total cost.
In the long run, AI may become similar to industries like electricity or cloud computing: prices gradually fall as competition increases, and the biggest winners are not necessarily those with the lowest advertised prices, but those with the most efficient infrastructure, the best software optimization, and the strongest customer relationships. Frontier model developers such as OpenAI and Anthropic still appear well positioned today because they lead in model quality and efficiency, but as AI becomes more commoditized, operational excellence and scale may become more important than simply having the newest model.
