I recently attended a networking event at Fudan University on Web3 and AI. As we were talking there, an interesting question suddenly came to mind.
What exactly are these companies offering large-model APIs, such as Kimi and MiniMax, actually selling?
On the surface, they sell model capabilities—question answering, generation, reasoning, search, and tool calling. But if we set aside these packaging descriptions and look only at the most basic commercial action, you will notice something very similar to Web3: every time you call a model, the system deducts a portion of your tokens.
Looking back at public blockchains from this perspective, you will find that these two sets of things actually feel very similar.
When you transfer funds, call a contract, or deploy code on Ethereum, you must consume gas; when you initiate a transaction on Solana, you must also pay network fees. Put simply, no action on-chain is free: the system converts your behavior into a measurable consumption of resources and then translates that into the corresponding fee.
With this comparison, things become interesting.
One person in the Web3 world says, "I just sent a transaction and consumed 0.01 ETH." Another person in the AI world says, "I just called a model and consumed 100,000 tokens." The former sounds like the language of the blockchain world, while the latter sounds like the billing method of cloud computing or SaaS products. But if you look one layer deeper, what they are doing is actually very similar: both divide an underlying resource into the smallest units that can be calculated, consumed, and settled, and then sell those units to developers and users.
From this perspective, I think what many AI companies are doing today and what many public blockchain projects did in the past are actually connected in terms of commercial structure.
They are all selling tokens.
Of course, "token" here does not mean that AI companies have actually issued a coin that can be freely traded, listed on exchanges, and speculated on, as public blockchain projects do. That is not what I mean. I mean that they are all selling a standardized unit of resource invocation.
In essence, they are all selling the right to invoke
When you use Kimi, you are not directly buying "an article" or "an answer." What you are buying is the model's ability to process text, the occupation of the context window, the consumption of the reasoning process, and the frequency and quota of API calls. The platform simply divides these originally abstract things into individual tokens and then charges you according to consumption.
Doing things on a public blockchain is the same. You are not directly buying the four words “transfer successful”; rather, you are paying for the resources consumed by the network to complete a bookkeeping entry, verification, ordering, and state update. The only difference is that the on-chain world calls the unit of measurement for this resource consumption gas, which is ultimately paid in native tokens such as ETH and SOL.
So AI companies and public blockchain projects are indeed very similar at a very fundamental level: neither directly sells results; rather, they sell “the right to invoke underlying computing resources.”
This observation is actually quite important. Because over the past few years, the Web3 industry has been very fond of the “utility token” narrative, but frankly, many projects have explained it too vaguely, and in the end what remains in everyone’s mind is still “issuing tokens, listing on exchanges, pumping the price, and market making,” which instead obscures the most basic commercial logic.
Now that AI companies have emerged, they do not tell you grand narratives, nor do they talk about community consensus, network states, or open finance. They simply tell you honestly: how much you use, how much is deducted, billed by token.
As a result, it suddenly became clear to people.
It turns out that the most basic meaning of the so-called token is not necessarily some financial asset, nor does it necessarily have to correspond to a price curve. First and foremost, it can simply be a unit of measurement, a unit of settlement, and a voucher for resource invocation.
Similar in appearance, different in nature
But if the article stops here, it is not enough. Because the tokens sold by AI companies and the tokens sold by public blockchain projects, although similar in appearance, are actually not the same thing.
The core difference lies in the different structures of rights behind them.
The tokens sold to you by AI companies are essentially an internal billing unit within the platform. You top up, open an account, obtain an API key, and then consume quota according to the platform rules. What you possess is not an asset that can circulate freely, be transferred freely, and exist independently of the platform, but a right of use recognized by the platform.
How can you understand it? It is more like points in a game, or invocation quota in a cloud provider’s backend, or the balance in a membership system. Of course, this thing has value because it can be exchanged for services; but its value boundaries, usage rules, and price adjustments are basically all in the hands of the platform. The platform can raise prices, limit traffic, change plans, and redefine the billing methods for different models. What you are facing is essentially a centralized company.
But public blockchain tokens are different.
Things like ETH and SOL are not merely units of account within a system; they are themselves native assets of the network. They can be held by you, transferred, traded, staked, and used as collateral, and they can also exist independently of any particular invocation. In other words, a public-chain token is not simply a number used for back-end bookkeeping; it is itself an object of market trading.
This is precisely where the biggest difference between the two lies: an AI token is more like a quota within a platform, whereas a public-chain token is more like an asset within a network.
One primarily exists within contractual relationships, while the other primarily exists within protocol networks and market trading. Once this difference emerges, many questions become clear. Why is it that although both are called tokens, their regulatory logic, market logic, and user expectations are completely different? Because although they share the same word, the legal and commercial structures behind them are fundamentally not the same thing.
One is platform pricing, the other is network pricing
Looking further, there is a second difference: who sets the price is different.
Today, the token prices of AI companies are basically set by the companies themselves. Which model you invoke, how much the input costs, how much the output costs, how much long-context costs, how much tool invocation costs—these matters are all preconfigured by the platform in the back end. Whether users accept them or not, in essence this is still platform pricing.
But gas on a public chain is not simply a matter of the platform attaching a price tag. On-chain fees are jointly affected by network congestion, market supply and demand, user bids, and protocol mechanisms. If you are willing to bid a higher price, your transaction may be packaged faster; the busier the network and the scarcer the resources, the higher the fees may be.
Therefore, the billing logic of AI tokens is essentially enterprise pricing; the billing logic of public-chain tokens is essentially closer to pricing jointly determined by the protocol and the market.
To put it more plainly, AI companies are in the platform business, while public-chain projects are in the network business. The core of the former is operating rights and control rights; the core of the latter is rules, nodes, incentives, and consensus.
One is selling services, the other is sustaining a network
There is also a third difference: to whom exactly are you paying money?
The money you pay to Kimi and MiniMax ultimately enters the revenue system of that company. The company then uses that money to buy GPUs, bandwidth, data centers, and research and development—all of which are matters of the company's internal cost structure. But from the user's perspective, you are dealing with a clearly identified service provider. What you are buying is its model services, system stability, and API capabilities.
But when you pay gas on-chain, you are not simply paying a particular company. What lies behind on-chain fees is the execution, validation, ordering and security maintenance of the entire network. In other words, what you pay for is not merely an API call, but the operating costs and security mechanisms of a decentralized network.
Ultimately, when AI companies sell tokens, it is more like platform billing; when public blockchain projects sell tokens, it is more like network incentives.
Both are related to resource consumption, but they are not the same form of organization, nor the same kind of commercial relationship.
AI, in turn, has taught Web3 a lesson
As I write this, I actually think the most valuable aspect of this observation is not that "AI looks a lot like Web3," but that it can help us re-understand an old question: why do many Web3 projects ultimately fail, while everyone naturally finds AI companies' token billing reasonable?
The reason is simple.
Because AI companies' tokens correspond to a very clear resource object and a very clear reason to pay.
When you use a model, you are genuinely consuming computing power. When you run a long context, you are genuinely occupying window resources. When you do searches or make tool calls, you are genuinely increasing platform costs. Each additional call you make means the platform bears an additional portion of marginal cost.
So when the platform divides this consumption into tokens and charges by token, the logic is very smooth. Users understand at a glance: the money I spend corresponds to the resources I actually consume.
But the problem with many Web3 projects in the past was not that they called it a token, but that behind it there was simply no such strong real consumption scenario.
When many projects discuss their business model, their first thought is not why users would keep using and keep paying, but "how to build up the token." How to distribute it, how to make its price rise, how to whip up market sentiment, how to make everyone believe this thing will be worth more in the future. As for what exactly users are paying for, and what resource consumption this token actually corresponds to, that is often not clearly explained.
So the order of entrepreneurship is easily reversed.
First establish the use case, then discuss the token
I think this may be the biggest insight AI companies offer Web3 entrepreneurs.
Many Web3 projects start with the first thought being "how to build up the coin." They first design the token model, and first think about circulating supply, unlock schedules, exchange listings, market making, and community sentiment. But looking at how AI companies operate today, the correct order may be exactly the reverse.
First, get one thing clear: who exactly are users paying for?
Then, solidify one thing: why do users pay repeatedly?
Then look further: in this business, does there truly exist a resource consumption that can be partitioned, measured, and settled?
Once these three things are clear, then you decide: whether this unit of measurement should be made into a token, whether it should be put on-chain, whether it should be freely tradable, and whether it should be given asset attributes.
Rather than starting by asking how to issue a coin, how to get listed, and how to manage market capitalization.
To put it even more plainly, it is not about first building up the token and then looking for a use case; rather, it is about first establishing the use case and then deciding whether this unit of measurement should become a token.
Many projects have actually gotten this sequence backwards from day one.
Not everything is worth putting on-chain
This way, all the logic that follows falls into place.
Not everything is worth putting on-chain. Because for many businesses, a centralized platform is inherently more efficient, the contractual relationships are clearer, and there is simply no need to force them into an on-chain architecture.
Nor is every unit of account worth trading. Because many units of account are only suitable as internal settlement tools; once they are placed on the market for free trading, they instead distort the original logic of their use.
Still less should every token have a secondary market. Because once it becomes a secondary-market asset, the motives of users in buying it, the motives of the project team in operating it, and the logic by which the market views it all change.
In many cases, the best token is not the one that rises the most sharply, but the one that least requires you to watch its price every day, yet can be continuously consumed, continuously settled, and continuously repurchased in real business.
This statement, I think, is actually very important.
Because it pulls the value of a token back from the price curve to business consumption itself. A truly healthy token, if it truly needs to exist, should first be a stable, clear, and easily understood unit of account within a business system, and only then can it possibly develop asset attributes further. Conversely, if the most important function of a token from the outset is to pump its price, then it has most likely gone astray from day one.
In the end, it comes back to that same statement
So let us return to today's topic.
"AI Companies Have Not Issued Coins, But They Are All Selling Tokens"
This is of course a headline with a bit of impact, and it deliberately retains a slightly provocative flavor. But if the underlying logic is explained thoroughly, you will find that this statement is not playing word games.
What AI companies sell is not a coin in the crypto-industry sense. What they sell is a token that has been corporatized, productized, and contractualized.
What the Web3 world sells should likewise not be merely a coin that can rise and fall. Its truly valuable part should originally also be the native pricing capability for a certain resource, a certain network capability, or a certain right to change state.
It is just that, in recent years, many people have misrepresented the latter.
Now it is instead AI companies that, using the most straightforward product logic and the most realistic charging methods, have helped everyone explain this matter clearly again.
The so-called token has never been mysterious. It is essentially about cutting an underlying resource that was originally continuous, vague, and difficult to settle into units that can be precisely consumed, and then embedding it into a commercial system.
If this system is a platform, it is more like a quota.
If this system is a network, it is more like an asset.
If there is real consumption behind this system, it has a reason to exist.
If there is only a price story behind this system, then it most likely will not last long.
This may be exactly a very realistic reminder that AI has brought to Web3.
Stop always thinking about issuing tokens. First think clearly: what exactly are you selling?
Are you selling a story, or a resource?
Are you selling imagination, or the right of invocation?
Are you selling a financial illusion, or an underlying capability that truly exists, can be repeatedly consumed, repeatedly priced, and repeatedly repurchased?
Once this question is thought through, the answer to whether many projects should issue tokens, whether they should go on-chain, and whether they should be built as open networks becomes apparent.
So I find what AI companies are doing today quite interesting, even somewhat ironic.
They have not issued tokens.
But they sell tokens every day.
And they sell them in a way that looks more like a legitimate business than many projects that have issued tokens.
