Token is expensive, which makes people feel sad.

This is not only the voice of people who are currently obsessed with Vibe Coding, but even the major Silicon Valley companies that have been crazy about Tokenmaxxing have begun to open Token restrictions for their employees.

But in fact, a counter-intuitive point is that for students who are currently using AI subscriptions, the Token you are using has actually been subsidized by major AI manufacturers, and the maximum subsidy may even be 70 times the subscription fee!

What is more worrying is that both OpenAI and Anthropic, two AI leaders, have entered the IPO sprint stage. After the two companies go public,

Will it be like after the "subsidy war" in the Internet era, the remaining companies began to raise the unit price, so that the price of Token returns to rationality?

The good news is that this probably won't happen. Recently, Google Ventures founder Bill Maris asked a question on the All-in podcast:

If Google decides to cut the token price by another 80%, how will OpenAI and Anthropic respond?

Coincidentally, not long ago, the start-up team Agnes AI explained in detail the possible "Token Free Era" in a live broadcast with Geek Park.

So, will the price of Token rise or fall in the future? And what does this mean for people who are already addicted to AI?

Token subsidies are all gone

Why is it said that the current price of Token is actually not expensive?

Because at least on the AI subscription system, the current prices of various AI companies are already "fracture prices" after subsidies.

Recently, SemiAnalysis conducted a detailed evaluation of the actual consumption of Token values and subscription fees under the OpenAI and Anthropic subscription models.

SemiAnalysis did a simple but effective thing - actually using AI to complete various tasks under the subscription plans of various AI platforms, and then using API public pricing to back-calculate how much the tokens of these tasks are worth. The results are as follows:

Pay attention to a rule: the more expensive the package, the higher the subsidy multiple. This in itself shows that these high-end packages are not for making money - they are a kind of "reverse pricing", using the most aggressive losses to retain the heaviest users. Because heavy users are developers and corporate decision-makers, once they are bound to a certain platform, they will attract the entire team and entire product line behind them.

Why do you still do it when you are losing to this extent? The standard answer is: burn money first to gain scale, and then increase the price to gain back the blood once the scale is established. This is how the mobile Internet works - Didi and Uber subsidized taxi fares worth tens of billions of yuan, and taxi fares increased after the subsidies ended; Meituan subsidized countless takeaway meals, but delivery fees increased after the subsidies ended. There is a key premise for this logic: a lock-in effect is established during the subsidy period.

Didi can increase prices because drivers cannot do without the order flow on the platform, and passengers cannot do without the drivers on the platform. Meituan can increase prices because merchants cannot do without its traffic and distribution network. When the subsidy ends, users will have been "locked" in the ecosystem, and switching costs will be extremely high.

But there is one fundamental difference between the AI war and the Internet -Token has almost no lock-in effect.

If the price of Claude increases, developers can migrate API calls to GPT or Gemini within a day - the interfaces of each company are becoming more and more standardized, and many development frameworks even have built-in multi-model switching functions. It’s even simpler for ordinary users: just change the URL. Unlike taxi-hailing, AI has a network of local drivers, unlike take-out, which has a delivery system, and unlike social media, which has a chain of friends. Token is token, no matter who produces it, it is the same thing.

This means that once the subsidy stops, users can be lost instantly. Subsidies are not "building barriers", but more like "maintaining a heartbeat" - as long as someone bids lower, users will run away.

And that’s not counting a new variable that’s sending everyone’s bills spiraling out of control: AI Agent.

When you chat with ChatGPT, a conversation may consume thousands of tokens. But when you ask the AI ​​Agent to perform a complex task - write a piece of code and then automatically debug it, analyze a document of dozens of pages and then generate a report - in one round, the token consumption is 5 to 30 times than that of a normal conversation. Some developers have measured that on the $100 Claude Max plan, one Agent programming session can burn nearly $100 worth of tokens. Uber’s CTO recently revealed that the company burned through its entire 2026 AI budget in four months.

The question is, can this Token subsidy war continue? Who could possibly stand and see the last thing after a chaotic battle?

Bill Maris thinks the obvious answer is the traditional giants.

Token as a weapon

To understand the true cruelty of this subsidy war, we need to first see clearly a structural asymmetry - the sources of ammunition for the participating parties are completely different.

Google generates more than $300 billion in advertising revenue annually. This is not the money given by investors, nor the money burned in financing, but a money printing machine that runs automatically every day. Billions of people around the world open search engines, watch YouTube, and use Gmail every day, and advertising fees automatically flow into their accounts. It doesn’t need a roadshow, it doesn’t need to woo analysts, it doesn’t need to explain to anyone why it’s spending the money.

Google uses advertising profits to subsidize AI tokens, which is like a person who owns an oil well engaging in a price war at a gas station - his oil comes from his own land, while his opponent's oil is bought from a bank loan.

OpenAI and Anthropic are the ones who took out loans to buy oil.

OpenAI has raised more than US$180 billion in cumulative financing, and its latest valuation exceeds US$850 billion. Anthropic has raised more than $130 billion. This money comes from venture capital and strategic investors - they don't give money as charity, they expect these companies to go public and get huge returns when they exit.

The trouble really started after it went public. Going public means that the financial statements are open to the world. Every quarter, Wall Street analysts look at revenue, profits, user acquisition costs, and marginal costs. When they figure you're actually losing $70 for every $1 in subscription fees you receive - no amount of brilliant growth stories can support the stock price.

Bill Maris made this logic very straightforward on the podcast. His original words were: "If I were Google and decided to cut the token price by 80%, what would happen to the business models of OpenAI and Anthropic?"

The host asked what the probability was. Maris did not hesitate: "100%. Capital as a weapon, tokens as a weapon (Capital as a weapon, Tokens as a weapon)."

This is not analyst speculation. Bill Maris is the founder and CEO of Google Ventures and vice president of special projects at Google. He has incubated Waymo and Google X. Everyone present understood: this is not a hypothesis, this is how he has seen with his own eyes how Google fights.

The scenario he painted was simple: Google announced an 80% price cut for the Gemini API. What will corporate customers do? If the product was of similar quality - Gemini is already on par with Claude and GPT in many benchmarks - but the price was four-fifths cheaper, would you continue to use the more expensive one?

Maris himself gave the answer: "If you are a company and you can pay 80% less and buy basically the same product at Google and Gemini, why wouldn't you? Then the pressure on those companies will become very severe."

OpenAI and Anthropic have almost no symmetric countermeasures. They can't follow through on price cuts - there's no money printing press and every dollar is investor money. Nor can they rely on the technology gap to maintain a premium - the gap between the big models is shrinking rapidly, today you are three months ahead and three months later you are tied. It’s not like the iPhone vs. Nokia generational technological generation gap. The moat between the AI ​​models is more like a dam made of sand, which will overflow when the tide rises.

According to Bill’s narrative, Google has a great chance of winning, but in the world of AI, can Google really have a monopoly? Meta can open source a free model at any time, China has DeepSeek and Byte, and Amazon is pushing its own model. When you bring the token down to a bargain price, your competitors don’t disappear—they also lower their prices.

In the AI war, there may be no winner.

Token’s “infinite game”?

Even people who don’t know much about history will make the following judgment about the outcome of the current AI war:

The first is the "Internet service" script - the story of Didi and the story of Amazon: first subsidize, then monopolize, and then raise prices to harvest. In this scenario, today's price war is just the prologue. In the end, one or two winners will occupy the vast majority of the market and gain pricing power. If so, the current huge losses are a good investment - just like Amazon, which lost money for two decades and eventually became the dominant player in both e-commerce and cloud computing.

The second type is the "Water, Electricity and Coal" script. Token becomes a standardized basic resource, like electricity, bandwidth, and cloud storage. No one can maintain pricing power for a long time because the product differences are too small and the switching costs are too low. Competition pushes prices infinitely toward the cost line, and profit margins approach zero. Eventually, governments may step in to regulate—just as they did with electricity and telecommunications a hundred years ago.

The difference between the two scripts depends on one word:

Lock.

Didi can increase prices because passengers are locked into the driver network, and drivers are also locked into the order flow. Amazon can raise prices because merchants are locked into its logistics and traffic ecosystem.

The lock-in effect is the cornerstone of the "loss first, profit later" model.

But AI tokens - as has been demonstrated repeatedly before - have almost no lock-in. APIs are standardized and switching costs are approximately zero. The core conditions for the establishment of the first scenario do not exist in the token product.

If the second scenario, the endgame of "water, electricity and coal" infrastructure, is closer to reality, what we are witnessing is not a war that will eventually determine the winner, but a race of attrition with no endgame.

Wang Xing, founder of Meituan, once described this competitive state. His insight is that there is no concept of "winning" in some competitions. The goal of the players is not to beat their opponents, but to ensure that they stay at the table. Because as long as you are still on the poker table, you can continue to raise funds, recruit people, and iterate. Leaving the table is the only way to lose.

Using this framework to re-examine today’s AI landscape, many seemingly contradictory things suddenly become clear.

OpenAI’s latest valuation exceeds $800 billion, not because it takes so much money to train the model. It needs so much money to continue its price war. Financing is not about winning, it is about "qualifying to continue fighting."

Google is planning to reduce the token price by 80%, not to eliminate OpenAI and Anthropic. It is ensuring that it remains a core player in the AI ​​era - just as it has ensured that it will not be left behind in the mobile era through free Android.

Anthropic has raised the API pricing of its latest flagship model, Fable 5, to twice that of the previous generation—$10 per million tokens for input and $50 per million tokens for output—seemingly as a “price increase.” In fact, it is actively screening out enterprise customers who are willing to pay for high-end capabilities, because it knows in its heart that it cannot win the subsidy war on the consumer side against Google.

Each round of price wars will expand the scale of AI use. Expanding scale means more data, more scenarios, and more developers pouring into the ecosystem. This in turn makes the models stronger for all participants. Participants use the war itself to attract resources to upgrade themselves - this is not a zero-sum game of life and death, but a process in which everyone becomes stronger together through competition, but it is unlikely that anyone will make huge profits.

Does this sound like what the power industry will eventually look like?

140 years ago, both Edison and Westinghouse thought they were competing for a winner-take-all market. They spent all their money on the bet that "whoever defines the standard of electricity will own electricity." But the fate of electricity tells us a simple truth:

When a technology is important enough, common enough, and standardized enough, it no longer belongs to any one company. It belongs to infrastructure.

The competition in AI is, on the surface, Google vs. OpenAI vs. Anthropic. It is a competition of model capabilities and a competition of financing scale. But zoom out and here’s what this competition is really doing: It’s accelerating the push of AI to a level of infrastructure that no one company can monopolize.

When Bill Maris said "100% going to happen," he probably wasn't just predicting a Google price cut. He may be unconsciously predicting a larger trend - in the world of AI, tokens will eventually belong to no one. Just like no one "owns" electricity today.

For OpenAI and Anthropic, this means something disturbing: Even if the technology is leading, even if they raise huge amounts of money, the future of "making big money from AI" they are chasing may not exist from the beginning. What they are facing is not a temporary price war, but a structural fate - what they are trying to build may essentially be the next generation of water, electricity and roads.

For users, to some extent, this may be good news. Because as long as the Token subsidy war continues, people can still enjoy the "good deal" with a cost of US$20 and computing power of US$400.