Author｜Space Monkey

Tokens are expensive; burning them hurts.

This is not only the voice of those currently obsessed with Vibe Coding, but even the big tech companies in Silicon Valley that previously frantically promoted Tokenmaxxing have started to set token restrictions for their employees.

However, an unconventional point is that for those currently using AI subscriptions, the tokens you are using have already been subsidized by major AI companies, and the maximum subsidy could even be as high as 70 times the subscription fee!

What is even more worrying is that the two leading AI players, OpenAI and Anthropic, have entered the IPO sprint stage. Once the two companies go public,

will it resemble the “subsidy war” of the internet era where the remaining companies start raising prices, causing token prices to return to rational levels?

The good news is that this situation may not occur. Recently, Bill Maris, the founder of Google Ventures, raised a question on the All-in podcast:

If Google decides to cut token prices by another 80%, how would OpenAI and Anthropic respond?

Interestingly, not long ago, the startup team Agnes AI explained in detail the possible arrival of the “Token Free Era” during a live broadcast with Geek Park.

So, will token prices rise or fall in the future? And what does this mean for those who are already addicted to AI?

01 Token subsidies are burning out

Why do we say that token prices are actually not high right now?

Because at least in the AI subscription model, the current prices of various AI companies are already the "discounted prices" after subsidies.

Recently, SemiAnalysis conducted a detailed assessment comparing the actual token consumption value and subscription fees under the subscription models of OpenAI and Anthropic.

SemiAnalysis did a simple yet effective thing—actually used AI under the subscription plans across various AI platforms to complete various tasks, and then calculated the token value of these tasks based on public API pricing. The results are as follows:

Note a pattern: the more expensive the plan, the higher the subsidy multiple. This itself indicates that these high-end plans are not designed to make a profit—they are a form of "reverse pricing," using the most aggressive losses to retain the heaviest users. Heavy users are developers and business decision-makers; once they are locked into a platform, they will bring along their whole team and entire product line.

Why continue to operate at such a loss? The standard answer is: burn cash to gain scale, then raise prices to recover costs. The mobile internet played this way—Didi and Uber subsidized hundreds of billions in ride-hailing fees, and after subsidies ended, ride prices went up; Meituan subsidized countless takeaway meals, and after subsidies ended, delivery fees increased. The logic holds if there is a key premise: a locking effect is established during the subsidy period.

Didi can raise prices because drivers depend on the order flow on the platform, and passengers depend on the drivers on the platform. Meituan can raise prices because merchants rely on its traffic and delivery network. By the end of the subsidies, users are already "locked" into the ecosystem, and switching costs are extremely high.

However, the AI war has a fundamental difference from the internet—tokens have almost no locking effect.

If Claude raises prices, developers can migrate their API calls to GPT or Gemini within a day—each platform’s interfaces are becoming increasingly standardized, and many development frameworks even have built-in multi-model switching capabilities. For ordinary users, it’s even simpler: just change the website. AI is not like ride-hailing with local driver networks, nor is it like food delivery with a delivery system, nor like social media with friend relationship chains. Tokens are just tokens; no matter who produces them, they are all the same.

This means that once subsidies stop, users can instantly churn. Subsidies are not about "building barriers," but more about "maintaining a heartbeat"—as long as someone offers a lower price, the users will leave.

And this does not even account for a new variable that is making everyone's bills uncontrollable: AI Agents.

When you chat with ChatGPT, the tokens consumed per conversation may be several thousand. But when you ask an AI Agent to perform a complex task—writing a piece of code and then auto-debugging, analyzing a dozens-page document and then generating a report—the token consumption can be 5 to 30 times that of ordinary conversations. Some developers have tested that on the $100 Claude Max plan, a single Agent programming session can burn up nearly $100 worth of tokens. The CTO of Uber recently revealed that the company burned through its entire AI budget for 2026 in just four months.

The question is, can this token subsidy war continue? Who is likely to stand tall after the chaos?

Bill Maris believes the answer is clearly traditional giants.

02 Token as a weapon

To understand the true brutality of this subsidy war, one must first recognize a structural asymmetry—the sources of ammunition for the combatants are completely different.

Google earns over $300 billion annually from advertising. This money does not come from investors, nor is it burned through fundraising, but rather a constantly running money-printing machine. Billions of people around the world open search engines, watch YouTube, and use Gmail daily, and advertising fees automatically flow into accounts. It doesn’t require roadshows, doesn’t need to please analysts, and doesn’t need to explain to anyone why this money is being spent.

Google subsidizing AI tokens with ad profits is like someone with an oil well engaging in a price war with gas stations—his oil comes from his own land, while his competitors have borrowed money from banks to buy theirs.

OpenAI and Anthropic are those who have borrowed money to buy oil.

OpenAI has raised over $180 billion in funding, with a latest valuation exceeding $850 billion. Anthropic has raised over $130 billion. This money comes from venture capital and strategic investors—they are not giving money for charity; they expect these companies to go public and look forward to hefty returns upon exit.

And after going public, the real trouble begins. Going public means financial statements are disclosed to the world. Every quarter, Wall Street analysts will scrutinize revenue, profit, customer acquisition costs, marginal costs. When they calculate that you are actually losing $70 for every $1 you receive in subscription fees—no matter how splendid the growth story, the stock price cannot hold up.

Bill Maris articulated this logic clearly on the podcast. His exact words were: “If I were Google, decided to arbitrarily cut token prices by 80%, what would happen to OpenAI and Anthropic's business models?”

The host pressed for the probability of that happening. Maris did not hesitate: “100%. Capital as a weapon, tokens as a weapon.”

This isn't an analyst's speculation. Bill Maris is the founder and CEO of Google Ventures, and also the Vice President of Google Special Projects. He has seen how Google fights wars firsthand.

He depicted a simple scenario: if Google announces an 80% price cut for the Gemini API, how will corporate clients respond? If the product quality is comparable—in many benchmarks Gemini has already matched Claude and GPT—but at a price 80% cheaper, would you continue to use the more expensive option?

Maris himself provided the answer: “If you are a company that can pay 80% less for a basically similar product at Google and Gemini, why wouldn’t you? Then the pressure on those companies would become extremely severe.”

Yet OpenAI and Anthropic have almost no symmetrical countermeasures. They cannot lower prices in response—there is no money printing machine; every dollar is investor money. They also cannot maintain a premium based on technological differences—the gap between large models is rapidly shrinking; today you may lead by three months, but three months later, others will catch up. This is not like the generational technological gap that existed between the iPhone and Nokia. The moats between AI models more resemble a dam built of sand, which gets washed away by a rising tide.

Under Bill's narrative, Google has a high chance of winning, but can Google really monopolize in the world of AI? Meta can open source a free model at any time, China has DeepSeek and ByteDance, and Amazon is pushing its own models. When you drive token prices down to rock bottom, competitors do not disappear—they also lower their prices.

In the AI war, there may be no winners.

03 The “infinite game” of tokens?

Even those who are not very clear about history can somewhat judge the current AI war’s ending as follows:

The first scenario is the “internet service” script—stories of Didi and Amazon: first subsidize, then monopolize, and finally raise prices to harvest profits. In this script, today's price war is just a prologue, and eventually, one or two winners will occupy the vast majority of the market and gain pricing power. If this is the case, the current massive losses are a worthwhile investment—just like Amazon lost for twenty years and eventually became a dual powerhouse in e-commerce and cloud computing.

The second scenario is the “electricity, water and coal” script. Tokens become a standardized basic resource, like electricity, bandwidth, and cloud storage. No one can maintain pricing power for long because the product differentiation is too small, and the switching costs are too low. Competition drives prices down to the cost line, and profit margins approach zero. In the end, the government may intervene to regulate—just like it did with electricity and telecommunications a hundred years ago.

The distinction between the two scripts depends on one word:

Lock-in.

Didi can raise prices because passengers are locked into the driver network, and drivers are locked into order flows. Amazon can raise prices because merchants are locked into its logistics and traffic ecosystems.

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

But AI tokens— as repeatedly argued above—almost have no lock-in. The API standardization and switching costs are nearly zero. The core condition for the first scenario to hold does not exist in the token product.

If the second scenario, the “infrastructure of water, electricity, and coal” is more realistic, what we are witnessing is not a war that will ultimately lead to winners and losers, but an endless war of consumption.

Wang Xing, the founder of Meituan, once described this competitive state. His insight is that some competitions do not have a concept of winning. The participants' goal is not to defeat their opponents but to ensure they always remain at the table. Because as long as you are still at the table, you can continue financing, hiring, and iterating. Leaving the table is the only way to lose.

Revisiting today's AI landscape through this framework makes many seemingly contradictory things suddenly clear.

OpenAI's latest valuation exceeds $800 billion, not because training models require so much money. It needs that much money to continue the price war. Fundraising is not for winning; it is for “being qualified to keep playing.”

Google's plan to cut token prices by 80% is not to eliminate OpenAI and Anthropic. It is to ensure its position as a core player in the AI age—just as it once ensured it would not be left behind in the mobile era with free Android.

Meanwhile, Anthropic raised the API pricing of its latest flagship model Fable 5 to double that of its predecessor—charging $10 per million input tokens and $50 per million output tokens—seemingly a “price increase,” but in reality, it actively selects enterprise clients willing to pay for high-end capabilities, because it understands that the consumer-end subsidy war cannot win against Google.

Every round of price wars will expand the scale of AI usage. Increased scale means more data, more scenarios, and more developers flowing into the ecosystem. In turn, this strengthens all participants’ models. Combatants attract resources to upgrade themselves through war itself—this is not a zero-sum game of you die, I live, but a process where everyone becomes stronger through competition, yet none are likely to reap huge profits.

Does this sound like the eventual state of the electricity industry?

140 years ago, Edison and Westinghouse thought they were vying for a winner-takes-all market. They bet their entire fortunes on the premise 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, generalized enough, and standardized enough, it no longer belongs to any one company. It belongs to infrastructure.

The competition in AI, on the surface, appears to be Google versus OpenAI versus Anthropic, a contest of model capabilities, a contest of funding scales. But zooming out, the real function of this competition is: it is accelerating AI towards a level of infrastructure that no company can monopolize.

When Bill Maris says "100% will happen," he might not just be predicting that Google will lower prices. He may be unconsciously predicting a larger trend—that in the world of AI, tokens will ultimately belong to no one. Just like today, no one "owns" electricity.

For OpenAI and Anthropic, this represents an unsettling reality: even with technological leadership and massive amounts of funding, the future they pursue of "making big money through AI" may have never existed from the start. They are facing not a temporary price war but a structural destiny—they are striving to build something that may essentially be the next generation of water, electricity, and infrastructure.

For users, in some sense, this may be good news. Because as long as the token subsidy war continues, people can still enjoy a “good business” of $20 costs and $400 computing power.