AI companies are not issuing their own tokens, but they are all frantically selling tokens.

Author: Liu Honglin

Today I attended a discussion event on Web3 and AI at Fudan University. While chatting on-site, an interesting question suddenly popped into my mind.

What exactly are companies that provide large model APIs, like Kimi and MiniMax, selling?

On the surface, they are selling model capabilities, offering services like Q&A, generation, reasoning, searching, and tool invocation. But if we set aside these superficial descriptions and look only at the most basic commercial actions, you’ll find something very similar to Web3:

Every time you invoke a model, the system deducts a portion of your tokens.

When you look back at public chains, you’ll realize that these two systems are actually quite alike.

In the world of Web3, one might say, “I made a transaction costing 0.01 ETH.” Meanwhile, in the AI world, someone might say, “I invoked a model, consuming 100,000 tokens.”

The former sounds like language from the blockchain world, while the latter sounds like the billing method of cloud computing or SaaS products. But if you look a little deeper, you’ll find they are doing very similar things:

Both slice a type of underlying resource into the smallest units that can be computed, consumed, and settled, and then sell them to developers and users.

From this perspective, I believe that many of the actions taken by AI companies today are actually quite similar in commercial structure to those of many public chain projects in the past.

They are both selling tokens.

Of course, this “token” does not mean that AI companies are issuing a coin that can be freely traded, listed on exchanges, or speculated upon like public chain projects. What I mean is that they are selling a standardized unit of resource invocation.

01 Essentially selling invocation rights

When you use Kimi, you are not directly buying “an article” or “an answer.” You are purchasing the model’s ability to process text, the usage of the context window, the consumption of the reasoning process, and the frequency and limits of API calls. The platform has simply broken down these originally abstract concepts into tokens and charges you based on consumption.

When you operate on a public chain, it’s the same. You are not directly buying the phrase “transfer successful,” but paying for the resources consumed to complete a transaction, validate, sort, and update the state of the network. The blockchain world just calls this unit of resource consumption “gas,” which is ultimately paid with native tokens like ETH or SOL.

Thus, AI companies and public chain projects, in a very fundamental sense, are indeed quite similar: they are not directly selling results but rather “the right to invoke underlying computational resources.”

02 Similar appearance, different nature

However, if the article stopped here, it wouldn’t be enough. Because while the tokens sold by AI companies and those sold by public chain projects may look similar, they are fundamentally different.

The core difference lies in the different rights structures behind them.

The tokens sold by AI companies are essentially a billing unit within the platform. You recharge, open an account, obtain an API key, and then consume credits according to the platform’s rules. What you hold is not an asset that can circulate freely, be transferred, or exist independently of the platform, but a type of usage right recognized by the platform.

How can you understand this? It’s more like point vouchers in a game, or the invocation limits in a cloud provider’s backend, or the balance in a membership system. This thing certainly has value because it can be exchanged for services; however, its value boundaries, usage rules, and price adjustments are essentially controlled by the platform.

In contrast, public chain tokens are different. ETH, SOL, and similar tokens are not just measuring units within the system; they are also native assets within the network. They can be held, transferred, traded, staked, mortgaged, and can exist independently of any specific invocation action.

03 One is platform pricing, the other is network pricing

The prices of AI company tokens today are essentially set by the companies themselves. What model you invoke, how much you input, how much you output, how much a long context costs, how much a tool invocation costs—these are all predetermined by the platform. Whether users accept this or not, it ultimately remains platform pricing.

Public chain gas, however, is influenced by more than just simple platform pricing. The costs on-chain are affected by network congestion, market supply and demand, user bids, and protocol mechanisms.

The billing logic for AI tokens is essentially corporate pricing; the billing logic for public chain tokens is closer to a joint pricing model between the protocol and the market.

04 AI teaches Web3 a lesson

The most valuable aspect of this observation is not that “AI is very similar to Web3,” but that it can help us re-understand an old question: why do many Web3 projects ultimately fail, while the token billing of AI companies seems inherently reasonable?

The reason is simple.

AI company tokens are backed by very clear resource objects and very clear justifications for payment.

When you use a model, you are genuinely consuming computational power. Running long contexts truly occupies window resources. When you conduct searches or make tool invocations, you are genuinely increasing the platform’s costs. Each additional invocation increases the platform’s marginal costs.

Thus, when the platform slices this consumption into tokens and charges based on tokens, the logic is smooth. Users can easily understand: the money they spend corresponds to the resources they actually consume.

However, the problem with many past Web3 projects was not that they called it a token, but rather that there was not a strong enough real consumption scenario behind them. When discussing business models, many projects first considered not why users would continue to use and pay, but “how to create a token.”

05 Clarify the use case before discussing tokens

I think the greatest inspiration AI companies offer to Web3 entrepreneurs may lie here.

First, clarify this: who are users actually paying for?

Then, solidify this: why would users pay repeatedly?

Next, look deeper: does this business truly possess a resource consumption that can be segmented, measured, and settled?

Instead of immediately asking how to issue tokens, list them on exchanges, or manage market capitalization.

06 Not everything is worth putting on-chain

Not everything is worth putting on-chain. Many businesses are simply more efficient on centralized platforms, where contractual relationships are clearer, and there is no need to force them into an on-chain structure.

Not every measurement unit is worth trading either. Many measurement units are only suitable as internal settlement tools; once they are freely traded in the market, they can distort the original usage logic.

Often, the best tokens are not those that increase in value the most, but those that do not require you to monitor prices daily while continuously being consumed, settled, and repurchased in real business.

07 Conclusion

So returning to today’s topic: “AI companies haven’t issued coins, but they are all crazily selling tokens.”

What AI companies are selling is not coins in the crypto sense. They are selling a type of token that has been corporatized, productized, and contracted.

What the Web3 world should sell is not just a coin that can rise or fall. The truly valuable part should be its native pricing capability for certain resources, network capabilities, or rights to state changes.

So, let’s not always think about issuing tokens. First, clarify what you are actually selling.

Are you selling a story or a resource?

Are you selling imagination or invocation rights?

Are you selling a financial illusion or a real capability that can be repeatedly consumed, priced, and repurchased?

This may be the most valuable reminder that AI brings to Web3.