AI Tokens = Cabbage?
My last article on the Chinese IT sector picked up on this pattern: usage of the science fiction term 降维打击 ("dimensionality-reduction attack") to describe overwhelming technological superiority crushing all competitors. Today's article takes a similar approach with a pricing-focused term: when markets converge on "price of cabbage" or 白菜价 (báicài jià) for a given product.
Context
The trigger for the article is the rising discipline of artificial intelligence tokenomics. Briefly: large language models (LLMs) receive inputs (your prompts) and produce outputs (responses) via word fragments known as tokens.
The global capability to produce tokens in LLM inference is finite and therefore must have a cost. This capability is constantly expanding via data center buildup, but token consumption is skyrocketing as well. Generally, the more tokens you consume, the more you pay. An LLM that provides the same result as another by spending fewer tokens is a more efficient model.
This means that the pricing of tokens (tokenomics!) is highly dynamic. The price that AI providers charge for tokens is a function of electricity costs, model training expenditures, competitive considerations, venture capital subsidies, and other variables.
Chinese AI Labs Choose Open Source/Weights
So where do the Chinese come in? Chinese AI labs' policy in the current AI era (6-12 months, as AI eras can't last longer) is mostly to open-source their models. This applies even to flagship or "frontier" LLMs. There are exceptions (MiniMax licenses in Europe, Doubao being closed source, some Qwen variants, etc) but the pattern holds. In fact, Xi Jinping graduated this emergent trend into a national policy at the July World AI Conference in Shanghai - less than a month ago at the time of writing.
Open-source LLMs can be deployed for free, meaning tokens can be extracted from the model for zero dollars. You still need powerful hardware to run them - and doing that does carry electricity and other costs. But you certainly skip eye-watering bills from US frontier labs (or even the Chinese labs themselves, who will optionally charge to run the model from their infrastructure). Perhaps the most striking factor is that Chinese AI labs give away much of their intellectual property even though training LLMs costs hundreds of millions of dollars in computing costs.
This is where I first saw the "price of cabbage" analogy among Chinese commentators. The expression pre-dates the AI revolution and tokenomic wars; I just came late to it. "Cabbage" is a proxy for "something very cheap". Therefore, driving high-tech products toward the price of cabbage can be defined as deploying overwhelming market forces to crush prices.
Chinese industry excels at crushing prices. It has already happened at all levels. In light industry it enables dirt-cheap (yet great quality) socks and t-shirts. In medium industry it happens with furniture and electronics (an iPhone is not priced like cabbage - but if it were manufactured elsewhere it would be much more expensive). We are watching this happen in heavy industries like automobiles (EVs) and ships. Beyond heavy industry lie digital services, where knowledge economies like the US and Europe have felt comfortable in recent decades as China dominated manufacturing.
The global price of inference tokens plummeting to cabbage levels would be bad news for some - mainly US frontier labs and their investors. It would be good news for others, starting with anyone who consumes AI tokens, including numerous US enterprises. If sustained indefinitely, the pattern would shift significant value and cost from the model layer of the global AI industry to others such as energy, GPUs/NPUs/TPUs, and applications that leverage the cabbage-priced models to serve users.
It's important to clarify that even if this were true, it won't happen linearly due to the dynamism and complexity of the industry. For example, US frontier labs are building or renting enormous data center capability that will reduce their token pricing. On the Chinese side, companies cut prices to compete with each other; "going to sea" 出海 is always secondary. In the ruthless domestic Chinese market, even 90%+ cuts in token prices are observable. But in 2026, Chinese labs have also increased token prices.
What Else Has Been Cabbage-ified?
As with dimension-reduction attacks, I conducted a brief survey of Chinese sources for uses of this expression. Here are my favorite examples:
1 - Some Chinese authors warn against attempting to drive sensitive industries such as lithography machines down to cabbage pricing. The reason? Crushed prices cripple profit margins, and without those the incentives for innovation and investment in entire industries would vanish. A temporary victory over consolidated players could turn Pyrrhic as everyone stagnates together.
This is one of my favorite US-China topics. In China, cutthroat competition and low pricing is considered desirable. It's culturally and even governmentally enforced. However, in excess it can lead to 内卷 (nèijuǎn) or destructive competition. In the US you have someone like Peter Thiel picking up on the exact same issue. In "Zero to One", Thiel bluntly says entrepreneurs should aim for monopolistic businesses, which are admittedly bad for societies but good for shareholders thanks to high margins. The lithography example shows at least some in China agree with Thiel.
2 - Cabbage-priced salaries. Apparently entry-level salaries at major Chinese tech companies are often described with this term (in contrast with special offers, bonuses, etc)
3 - Computer storage. During 2023, Chinese producers like YMTC cut prices by circa 50%, hitting the dominant US and South Korean players in the industry. In 2026 we are in a comparable market war pitting the same players around memory. However, RAM prices are rising rather than approaching cabbage levels.
4 - Ransomware. I was surprised by this one. Apparently some ransomware payloads targeting Chinese systems are demanding ridiculously low ransom payments. This would represent a return to the origins of ransomware, which targeted individual users before moving to the heyday of "Big Game Hunting" (large companies).
5 - Photovoltaic cells and lithium EV batteries. These are to be expected, as they are China's most prominently dominated export markets nowadays.
6 - I had a 2022 cutoff in my research, but found discussions referring to steel selling at cabbage prices circa 2015, and optical fiber pricing wars in 2018. I find the latter interesting because to this day it is commonplace to see quality complaints about Internet connectivity in China, both at the consumer and the business levels.
7 - As I was finishing this draft, the regime-operated Global Times had an article on polysilicon producers accounting for 90% of Chinese production agreed not to engage in 内卷 (competitive involution). The lessons from photovoltaic panels crumpling to cabbage prices are mentioned prominently. SAMR (the relevant regulator) is publicly tasked with ensuring cabbage-ification doesn't happen in additional industries. The term "cabbage" doesn't appear in the English version, though it might in the original.
Parting Thoughts
The Chinese cabbage pricing issue is a small insight into how China simultaneously practices capitalism and socialism. Extreme market dynamics, resulting in cabbage prices, can humble even the mightiest corporation. On this, hypercapitalists like Thiel and socialists like the Communist Party of China agree.
Hope you enjoyed this small cultural snippet. I produce a large volume of intelligence, cybersecurity governance, and other insights for customers - don't hesitate to reach out if you're interested. My companies also provide automated software to monitor for emerging issues of interest in the Chinese Internet.
AI Manifest for This Article
I believe full disclosure of AI usage in creative work is increasingly a must these days. So here is mine:
All ideas by me
All writing by me
Headline cabbage image generated via Gemini (with a clever prompt if I may say so!)
Draft submitted for review to Gemini and ChatGPT. Neither made substantial contributions (I would not have hesitated to incorporate some)
Programmatic queries for mentions of 白菜价 conducted via agentic AI