Subsidies → Token-based billing → Price cuts! OpenAI initiates a price war—Is a turning point in token economics approaching?

```

The commercial narrative of generative AI is undergoing its most profound self-examination in the past three years. From user acquisition by subsidies, hidden costs in monthly subscriptions, to token-based billing triggering enterprise bill crises, the AI industry has completed a three-level commercialization leap within three years—a potential price war may reset the entire monetization logic again.

According to The Wall Street Journal, OpenAI is considering a significant reduction in token fees charged to users, in order to compete for enterprise customers from rival Anthropic. Insiders say this move is partly a preemptive step, as OpenAI anticipates Anthropic will also initiate similar price cuts. OpenAI CEO Sam Altman recently acknowledged at an event that the cost of AI usage has become “a huge problem,” and stated they would “help people get more value with less spending.”

The timing of this news is particularly sensitive. OpenAI has just secretly submitted its IPO application this week, and Anthropic is also approaching its own IPO countdown. Meanwhile, Bloomberg's Silicon Data LLM Token Spend Index has declined for seven consecutive trading days, marking the longest drop since January this year, reflecting deep market anxiety over the sustainability of AI bills. The report bluntly states, a price war will directly erode the profit margins of both companies—while both are already losing billions of dollars due to the massive computing power required by AI systems.

The core of this discussion is no longer just a price cut decision, but a more fundamental question: When the narrative of “the more token consumption the better” comes to an end, who will tell the next commercial story in the AI industry, and how.

Three Stages: From Monthly Package Subsidies to Token Bills

The commercialization of generative AI has undergone three clear stages in just three years.

Stage One: Monthly and yearly subscriptions set the industry tone. In February 2023, OpenAI launched ChatGPT Plus at $19.99 per month, setting a precedent for paid large model C-end (consumer) services. Baidu, Alibaba, and Tencent soon followed, and fixed monthly subscriptions became the norm for initial business models.

Stage Two: The subsidy war breaks out fully. In order to boost ARR (Annual Recurring Revenue), a key anchor for fundraising valuation, companies turned to large-scale subsidies: Google offered 15 months of Gemini Advanced to students for free, OpenAI launched Team membership at $1 for the first month, Bytedance Doubao entered with pricing “99.3% lower than the industry average,” and Baidu announced its core model would be free. The nature of subsidies is trading losses for growth—reports say Microsoft lost over $20 per user per month on average under Github Copilot’s subscription model, with heavy users losing up to $80 monthly.

Stage Three: Forced switch to usage-based billing. On June 1, 2026, Microsoft announced all Github Copilot plans would switch to token-based billing, with the $19 monthly fee converting directly to token quota. This shift revealed the real costs long hidden under the subscription model—according to Reddit users, a single intelligent agent programming session can consume $30-40; a monthly package can be exhausted in a single use.

Out-of-Control Billing: When Tokens Become More Expensive Than Humans

The implementation of token-based billing exposes the true face of AI costs for enterprises.

The numbers on enterprise bills are staggering. Uber COO Andrew Macdonald publicly said in May 2026 that the line between increased token consumption and actual product improvement “doesn’t exist,” and even coined a word: “tokenmaxxing,” describing employees performing valueless tasks just to run up usage.

More directly: Uber exhausted its annual token budget in the first four months of 2026; Salesforce estimates it will pay Anthropic around $300 million for the year.

Anthropic’s own developer documentation shows developers using Claude Code incur an average cost of about $13 per workday, with 90% incurring less than $30 a day—meaning a 10-person development team could spend over $75,600 a year on token fees alone.

The input-output ratio is equally alarming. Enterprise data platform Entelligence.AI, after aggregating data from 2444 companies, found that for every $1 spent on AI tokens, only 18 cents generated actual user-facing value; 44 cents went to fixing bugs introduced by the AI, 27 cents to rework, and 11 cents on review friction.

Facing spiraling bills, enterprises have started to manage usage proactively. Amazon halted its internal AI usage leaderboards, urging employees “not to use AI for AI’s sake.” Microsoft plans to phase out Claude Code subscriptions for some key product teams. Goldman Sachs notes that some enterprises’ AI token expenses now account for 10% of total employee labor costs, and this could rise further in coming quarters. This isn’t the disappearance of demand, but the end of the rough-spending era of AI.

Act Four: The Price War Begins, OpenAI Considers Major Price Cut

It’s against this backdrop that the fuse of a price war is lit.

According to The Wall Street Journal, Altman’s consideration of price cuts was directly triggered by pressure to catch up with Anthropic. Anthropic’s revenues have surged recently, its coding tool Claude Code is gaining fans among software engineers, and this five-year-old startup’s valuation has even surpassed OpenAI’s for the first time.

However, the cost of this price war will be exceptionally heavy. A major price cut would further squeeze both companies’ already negative profit margins, with little room to maneuver in the competitive landscape.

And investors have long recognized a fundamental risk: OpenAI and Anthropic’s products are highly substitutable, clients can easily switch from one to the other—meaning price cuts may retain customers short-term but offer no real moat, merely delaying the loss in market share.

This dilemma is also transmitted outward by the financial ties between cloud giants and AI labs.

According to The Information’s summary of disclosed corporate documents, OpenAI and Anthropic together account for more than half of Microsoft, Oracle, Google, and Amazon’s $2 trillion in future cloud service commitments. If price cuts force down revenue forecasts, the chain will face two-way pressure.

US neuroscientist and AI expert Gary Marcus said: “This further exposes OpenAI’s vulnerability and how serious its predicament is. Once OpenAI goes into decline, it could drag down Nvidia, Oracle, Coreweave, and others. The situation is deteriorating rapidly.”

Bulls and bears are openly at odds on Wall Street. JPMorgan TMT analyst Mark Schilsky believes current billing anxiety is merely “the smallest speed bump on the path to higher spending”: even if per-million-token prices drop, as AI adoption rises among US companies, total token usage will inevitably soar; plus, agentic AI pushes single-task token consumption to multiples of traditional Q&A usage, so in the long run, total spending will exceed current levels.

Goldman Sachs semiconductor analyst Jim Covello takes a more pessimistic view, arguing that current industry prosperity funnels almost all value to semiconductor companies, a “historically unprecedented and unsustainable” phenomenon. Once companies face the true price of pay-as-you-go, capital flows supporting GPU procurement and model training may reverse.

Act Five: The Next Story for Token Economics?

After the price war, the next chapter in the commercialization of AI has yet to be written, but the outline is emerging.

Citadel Securities’ report offers one directional framework: tiered billing and scarcity-based pricing. The core logic is that inference-intensive frontier AI won’t disappear, but will be increasingly concentrated among large enterprises able to shoulder compute costs; for the broader set of companies, until physical constraints ease, simpler models may be more productive. This means AI usage will become tiered—frontier models for high-value, complex tasks, and cheap/local models for routine/batch jobs.

JPMorgan takes a relatively optimistic view: even as the unit price per token decreases, the rise of agentic AI (agent-based AI) will multiply token consumption per task—existing data shows that agentification raises per-task token usage to 3.5 times the prior level—so total outlays could still expand, and today’s billing anxiety might be merely “the smallest speed bump on the way to higher spending.”

Nebius Chief Revenue Officer Marc Boroditsky proposed “valuemaxxing”—advocating that the industry shift from maximizing token consumption to ensuring every token truly generates value. This is becoming industry consensus—but real business adoption still requires AI labs to establish a pricing system that reflects real costs and is also acceptable to enterprise clients, which remains the core unresolved issue in all debates today.

However, in this price war, the most overlooked variable may be Chinese models.

According to June data from US enterprise expense platform Ramp, DeepSeek has become the fastest-growing software subscription among US enterprises. Ramp’s chief economist Ara Kharazian particularly emphasized this is not local deployment of open-source models, but “enterprises are sending and receiving data directly to DeepSeek,” i.e., paid, direct usage—he admits “didn’t expect US firms would use DeepSeek.” Third-party estimates put DeepSeek V4-Pro API’s average price at about one-tenth of GPT-5.5, and one-eleventh that of Claude Opus 4.7.

While OpenAI and Anthropic fight, the ultimate beneficiary may be the player whose DNA already contains “inclusive pricing,” and who doesn’t need to account for profitability to IPO investors. This may not be the most popular outcome of the price war, but is becoming a reality that can no longer be ignored.

Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it consider individual users’ special investment objectives, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article accord with their circumstances. Investment based on this content is at your own risk. ```