Is the “Token Subsidy War” Among AI Giants Coming to an End?
TechFlow Selected TechFlow Selected
Is the “Token Subsidy War” Among AI Giants Coming to an End?
It can be further reduced by 80%.
Navigating Web3 tides with focused insightsAuthor | Yu Hangyuan
Tokens are expensive—so expensive they hurt.
This isn’t just the sentiment of those obsessed with “vibe coding.” Even Silicon Valley tech giants that once enthusiastically championed “tokenmaxxing” have begun imposing token usage limits on their own employees.
Yet here’s a counterintuitive truth: students currently using AI subscription services are already benefiting from massive subsidies by major AI companies—subsidies so generous they may reach up to 70 times the subscription fee!
Even more worrying is the fact that OpenAI and Anthropic—the two leading AI companies—have both entered the final sprint toward IPO. Once these firms go public, will we see a repeat of the “subsidy wars” of the internet era, where surviving players raise prices en masse, pushing token costs back toward rational levels?
The good news is that this scenario may never materialize. Recently, Bill Maris, founder of Google Ventures, posed a provocative question on the All-in podcast:
“What would happen if Google slashed token prices by another 80%—how would OpenAI and Anthropic respond?”
Coincidentally, not long ago, startup Agnes AI, during a livestream with GeekPark, laid out in detail the potential arrival of a “token-free era.”
So—will token prices rise or fall in the future? And what does that mean for those already addicted to AI?
01 Token Subsidies Are Already Smoking Hot
Why do we say tokens aren’t actually expensive today?
Because—at least under current AI subscription models—the prices offered by AI companies are already deeply subsidized “fire-sale” rates.
Recently, SemiAnalysis conducted an in-depth analysis comparing actual token consumption value against subscription fees for OpenAI and Anthropic.
SemiAnalysis did something simple yet powerful: it used various AI platforms’ subscription plans to complete real-world tasks, then reverse-calculated the monetary value of consumed tokens based on publicly disclosed API pricing. The results are as follows:
Notice one pattern: the more expensive the plan, the higher the subsidy multiplier. This alone reveals that premium plans aren’t designed for profit—they’re examples of “reverse pricing,” deliberately losing money at scale to retain the heaviest users. Why? Because heavy users are developers and enterprise decision-makers. Once locked into a platform, they bring entire teams—and entire product lines—with them.
So why sustain such staggering losses? The textbook answer is: burn cash now to gain scale, then raise prices later to recoup losses. That’s exactly how mobile internet played out—Didi and Uber subsidized billions of RMB in ride-hailing fares; after subsidies ended, fares rose. Meituan subsidized countless food deliveries; afterward, delivery fees increased. For this logic to hold, one key condition must be met: lock-in effect must be established during the subsidy period.
Didi could raise prices because drivers couldn’t survive without its order flow, and passengers couldn’t live without its driver network. Meituan could raise prices because merchants depended on its traffic and logistics infrastructure. When subsidies ended, users were “locked in”—switching costs were prohibitively high.
But the AI arms race differs fundamentally from the internet era—tokens carry almost no lock-in effect.
If Claude raises prices, developers can migrate their API calls to GPT or Gemini within a day—interfaces across providers are increasingly standardized, and many development frameworks even natively support multi-model switching. For ordinary users, it’s even simpler: just switch websites. Unlike ride-hailing (with local driver networks), food delivery (with logistics systems), or social media (with friend networks), AI has no inherent network effects. Tokens are tokens—regardless of origin, they’re functionally identical.
That means users can flee instantly once subsidies stop. Subsidies don’t build moats—they merely sustain vital signs. As soon as someone offers lower prices, users vanish.
And this doesn’t yet account for a new variable threatening to blow everyone’s budgets: AI agents.
A single ChatGPT chat might consume only a few thousand tokens. But when you deploy an AI agent to execute complex tasks—writing code and automatically debugging it, analyzing dozens of pages of documents and generating reports—token consumption surges to 5–30 times that of normal conversation. Developers have tested this: a single agent-powered coding session on Claude Max—a $100/month plan—can burn through nearly $100 worth of tokens. Uber’s CTO recently revealed the company exhausted its entire 2026 AI budget in just four months.
So—the question remains: can this token subsidy war continue indefinitely? Who stands the best chance of emerging victorious from this chaotic brawl?
Bill Maris believes the answer is clearly traditional tech giants.
02 Token as a Weapon
To grasp the true brutality of this subsidy war, we must first recognize a structural asymmetry—the combatants’ ammunition sources differ radically.
Google generates over $300 billion annually in advertising revenue. This isn’t investor capital or venture funding—it’s a self-sustaining, daily-printing cash machine. Billions of people worldwide open search engines, watch YouTube, use Gmail—ad revenue flows in automatically. Google doesn’t need roadshows, analyst appeasement, or justification for spending.
Subsidizing AI tokens with ad profits is like an oil tycoon waging a gas station price war—his oil gushes from his own land, while rivals must borrow from banks to buy theirs.
OpenAI and Anthropic are precisely those borrowing to buy oil.
OpenAI has raised over $180 billion in cumulative funding, with a latest valuation exceeding $85 billion. Anthropic has secured over $13 billion. These funds come from venture capitalists and strategic investors—not charity. They expect IPO exits and substantial returns.
But going public marks the real beginning of trouble. Public listing means full financial transparency. Every quarter, Wall Street analysts scrutinize revenue, profit, user acquisition cost, marginal cost. When they calculate that every $1 in subscription revenue incurs $70 in actual loss—the most dazzling growth story won’t prop up the stock price.
Maris spelled out this logic plainly on the podcast: “If I were Google and decided to slash token prices by 80% arbitrarily—what happens to OpenAI and Anthropic’s business model?”
The host asked how likely that was. Maris didn’t hesitate: “100%. Capital as a weapon, tokens as a weapon.”
This isn’t analyst speculation. Bill Maris founded and served as CEO of Google Ventures, and also held the role of VP of Special Projects at Google—he incubated Waymo and Google X. Everyone present understood: this wasn’t hypothetical. It was firsthand knowledge of how Google fights.
His scenario is straightforward: Google announces an 80% price cut for the Gemini API. What do enterprise customers do? If quality is comparable—which benchmarks show it already is, matching Claude and GPT—yet the price drops by 80%, would you stick with the pricier option?
Maris answered himself: “If you’re a company, and Google/Gemini lets you pay 80% less for essentially identical output—why wouldn’t you? Then pressure on those other companies becomes extremely severe.”
OpenAI and Anthropic possess virtually no symmetrical countermeasures. They can’t match the price cut—no printing press, every dollar comes from investors. Nor can they sustain premium pricing via technical advantage—the gap between large language models is narrowing rapidly. A three-month lead today vanishes in three months. This isn’t iPhone vs. Nokia—a generational tech gap. AI model moats resemble sandcastles—tides wash them away instantly.
In Maris’s narrative, Google holds strong odds—but can Google truly monopolize AI? Meta could open-source a free model anytime; China boasts DeepSeek and ByteDance; Amazon pushes its own models. Once tokens hit cabbage prices, competitors don’t vanish—they slash prices too.
The AI arms race may produce no winners.
03 The “Infinite Game” of Tokens?
Even those unfamiliar with history instinctively foresee two possible endings to today’s AI arms race:
First, the “internet services” script—Didi’s story, Amazon’s story: subsidize first, monopolize next, then raise prices and harvest profits. Under this script, today’s price war is merely the opening act—eventually one or two winners dominate the market and control pricing. If so, today’s massive losses are sound investments—like Amazon’s two-decade losses before becoming e-commerce and cloud computing king.
Second, the “utilities” script—tokens become standardized infrastructure, like electricity, bandwidth, or cloud storage. No firm can sustain pricing power long, because products are near-identical and switching costs near-zero. Competition drives prices relentlessly toward marginal cost, margins approaching zero. Eventually, governments may intervene—just as they did with electricity and telecom a century ago.
The divergence between these two scripts hinges on one word:
Lock-in.
Didi raised prices because riders were locked into its driver network, and drivers into its order flow. Amazon raised prices because merchants were locked into its logistics and traffic ecosystem.
Lock-in is the bedrock of the “lose-now-win-later” model.
But AI tokens—as repeatedly argued above—lack meaningful lock-in. APIs are standardized; switching costs approach zero. The core prerequisite for Script One simply doesn’t exist for tokens.
If Script Two—the utilities-infrastructure endpoint—is closer to reality, then what we’re witnessing isn’t a war destined for decisive victory, but an endless attrition contest.
Meituan founder Wang Xing once described this competitive state. His insight: some competitions lack a concept of “winning.” Participants aim not to defeat rivals, but to stay seated at the table. As long as you remain in play, you keep raising funds, hiring talent, iterating. Leaving the table is the only true loss.
Applying this lens to today’s AI landscape suddenly clarifies many seemingly contradictory developments.
OpenAI’s latest valuation exceeding $80 billion isn’t because training models demands that much capital. It needs that much money to keep fighting price wars. Fundraising isn’t about winning—it’s about staying eligible to fight.
Google preparing to slash token prices by 80% isn’t about eliminating OpenAI and Anthropic. It’s about ensuring Google remains a central player in the AI era—just as it guaranteed relevance in mobile through free Android.
Meanwhile, Anthropic’s latest flagship model Fable 5 charges double its predecessor’s API rate—$10 per million input tokens, $50 per million output tokens. Though seemingly a price hike, it’s actually deliberate customer segmentation: targeting enterprises willing to pay for premium capabilities. Anthropic knows it can’t outspend Google in consumer-level subsidy wars.
Each round of price war expands AI adoption. Scale brings more data, more use cases, more developers into the ecosystem—making every participant’s models stronger. Combatants upgrade themselves by waging war itself—not a zero-sum death match, but a collective upward spiral where all grow stronger, yet none reap windfall profits.
Doesn’t this sound eerily like the electricity industry’s ultimate shape?
140 years ago, Edison and Westinghouse believed they were battling for a winner-takes-all market. They bet their entire fortunes on “who defines the standard owns electricity.” Yet electricity’s fate teaches a simple lesson:
When a technology becomes sufficiently important, universal, and standardized, it ceases to belong to any single company. It becomes infrastructure.
On the surface, AI competition appears as Google vs. OpenAI vs. Anthropic—model capability showdowns, fundraising contests. Zoom out, however, and the real function of this competition emerges: accelerating AI’s ascent to infrastructure status—beyond any single company’s control.
When Bill Maris says “100% it will happen,” he may not just be predicting Google’s price cut. He may be unconsciously forecasting a broader trend—that in the AI world, tokens ultimately belong to no one. Just as no one “owns” electricity today.
For OpenAI and Anthropic, this implies an unsettling truth: even with technological leadership and astronomical funding, their envisioned future—“getting rich on AI”—may never exist. They aren’t fighting a temporary price war; they’re confronting a structural destiny—their life’s work may inherently become tomorrow’s water, electricity, and highways.
For users, however, this may be welcome news. As long as the token subsidy war continues, people will still enjoy “great deals”: $20 worth of cost delivering $400 worth of compute power.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
