Author: Ray Wang; Source: X,@wangray
A consensus seems to be gradually forming in the AI community: The AI industry = token economy. Strip away all the fancy narratives, and what's left are just two things: buying tokens and selling tokens.
This statement sounds very fundamental, very first-principles, and I thought it made perfect sense when I first heard it.
But the more I think about it, the more something seems off.
.
Tokens are not relations of production, but units of measurement
If you look at any society from a Marxist perspective, the first point of entry is never "what was produced," but who owns the means of production—landlords or peasants, capitalists or workers. From this perspective, everyone's position, interests, and fate are clear.
In an agricultural society, "growing and selling grain" is not a relation of production, but an act of labor. Landlords and tenant farmers both "grown grain" and "sold grain," but they are two completely different classes with diametrically opposed fates.
Stopping the analysis at "creating/selling tokens" is like stopping the analysis of an agricultural society at "growing/selling grain"—technically correct, but meaningless.
What is a token in AI? It's like "food," "yield per acre," a unit of measurement for consumption and output. In the electricity era, we didn't talk about "electricity economy," even though everything consumed electricity. In the internet era, we didn't talk about "data packet economy," even though everything was transmitted via TCP/IP. Why then, with AI, do we start talking about "token economy"? Because it sounds deeper, but essentially, it just stops the analysis at a level that "looks low-level, but isn't actually low-level enough." There are only four true means of production. If you dissect the entire AI industry chain, you'll find that what truly determines who makes money and who gets rented is not tokens, but four indivisible things: Computing power: GPUs, electricity, data centers, and chip manufacturing capabilities. Weights: The trained model itself. Data: Training data + continuously generated user behavior data. Distribution: The entry points for users to access AI: operating systems, browsers, super apps, and IDEs. Tokens are not on this list. Tokens are a "measurement of output" resulting from the collaboration of these four means of production, not the means of production themselves. Whoever holds at least one of these four is the landlord in the AI era. Everyone else is a tenant. A Class Analysis in the AI Era If you view the world through the lens of "token economics," you'll see two layers: those who sell tokens and those who buy tokens. But the real world has a seven-layer structure: **Computing Power Landlords:** NVIDIA, TSMC, the three major cloud computing companies, and energy giants. They control the "land." All tokens must be computed on their hardware. The initial funding for AI largely went into their pockets. **Power Aristocrats:** Anthropic, OpenAI, Google DeepMind. They possess "production capacity" but need to pay rent to computing power landlords. The stronger the model, the higher the rent.
**Data Lords:** Reddit, X, TikTok, Xiaohongshu, Bloomberg. They control the "raw materials for the next generation of power." No matter how strong the model, without new data, it will eventually run out of resources.
**Distribution Oligopolies:** Apple, WeChat, Chrome, Android. They control the "user entry point." All AI products ultimately reach users through them. They decide who is seen and who is ignored.
**Token Middle Class (Application Layer):** Notion AI, Perplexity, Cursor, etc. They're doing well, but profits have to be distributed upstream. Their ability to move upstream depends on whether they can accumulate data or control distribution. Token Shareholders (Shell Layer): Thousands upon thousands of wrappers. No computing power, no weight, no data, no distribution. Upstream changes in pricing cause collective harm. Token Self-Farmers: People running Llama on their own computers, doing local inference. Independent, free, no rent to pay, but minimal output, unable to shake any existing structure. The dichotomy of the "Token Economy" compresses these seven layers into two. Once flattened, it's impossible to see who will win and who will die. The real competition isn't about selling tokens, but about controlling the conditions. If you connect these four means of production, you'll find they're not independent, but form a closed loop: **Distribution → Data → Weighting → Usage → Back to Distribution.** Whoever can form this closed loop will have a long-term advantage. This is why true "landlords" often control multiple means of production simultaneously: Google: Distribution + Data + Weighting Meta: Data + Distribution + (Influencing weighting through open source) Apple: Distribution + Devices + (Currently adding weighting) Single-point capabilities are not barriers; closed-loop systems are. "Token economy" is the rhetoric of beneficiaries. Finally, let's discuss why "AI is a token economy" has become a consensus. Because this narrative is most beneficial to the true holders of the means of production. Drawing everyone's attention to the token layer—competing on who sells more, who sells at the highest price, and whose application has the most added value—means everyone is fighting each other in the mid-to-lower reaches, while the holders of the four means of production quietly collect rent at the top. NVIDIA doesn't care if you're doing legal AI or medical AI, as long as the tokens are processed on its cards. OpenAI doesn't care how fancy your shell is, as long as the APIs all go through it. Apple and WeChat don't care how cool your agent is, as long as users have to click through them. Thus, competition begins in the mid-to-lower reaches, while those who truly control the means of production steadily collect rent at the top. What Chairman Mao saw clearly about China in 1925 wasn't "what should peasants plant," but rather "why peasants are poor, who took their surplus, and how to reverse this relationship." Today, when discussing AI, the question shouldn't be, "What token should I sell, to whom, and at what price?" The question should be: Do I possess any of those four means of production? If not, can I acquire them? If not, how do I plan to survive? If you can't acquire them, stop deceiving yourself into thinking you're running an AI company; you're merely a tenant farmer in the AI era.
Weatherly
Anais