Subsidy → Token billing → Price reduction, OpenAI launches price war, is the turning point of Token economics approaching?

The commercialization narrative of generative AI is facing its most profound self-examination in three years. From exchanging subsidies for users and hiding the cost of monthly subscription to triggering a corporate billing crisis through token based billing, the AI industry has completed a three-level leap in commercialization in three years – and a potential price war may bring the entire monetization logic back to zero.

According to The Wall Street Journal, OpenAI is considering a significant reduction in token fees charged to users in order to compete for corporate clients from its competitor Anthropic. According to insiders, this move is partly to “seize the initiative”, and OpenAI expects Anthropic to take similar price reduction actions. OpenAI CEO Sam Altman recently admitted at an event that the cost of using AI has become a “huge issue” and stated that it will “help people get more value with less expenditure”.

The timing of this news is particularly sensitive. OpenAI has secretly submitted its IPO application this week, and Anthropic is also in the countdown stage to go public. At the same time, the Bloomberg Silicon Data LLM Token spending index has fallen for seven consecutive trading days, setting a record for the longest consecutive decline since January this year, reflecting the market’s deep anxiety about the sustainability of AI bills. The report bluntly states that the price war will directly erode the profit margins of both companies – and both companies are currently losing billions of dollars due to the massive computing power required for AI systems.

The core of this discussion is no longer just a price reduction decision, but a more fundamental question: when the narrative of “the more tokens consumed, the better” comes to an end, who will tell the next commercialization story of the AI industry, and how will it be told.

01
Initial three stages: from monthly package subsidies to token billing
The commercialization of generative AI has undergone three distinct stages of evolution in just three years.

In the first stage, monthly and annual subscription sets the tone for the industry. In February 2023, OpenAI launched ChatGPT Plus with a monthly fee of $19.99, setting a precedent for large model C-end payment; Baidu, Alibaba, and Tencent followed suit, and fixed monthly subscription became a standard feature of the primary business model.

In the second stage, the subsidy war broke out comprehensively. In order to raise ARR (annual recurring income), the core anchor of financing valuation, manufacturers turned to large-scale subsidies: Google provided 15 months of Gemini Advanced for students for free, OpenAI launched a team version member of $1 in the first month, ByteDance Doubao entered the market at a price of “99.3% lower than the industry price”, and Baidu announced that the core model was free. The essence of subsidies is to trade losses for growth – according to reports, Microsoft’s GitHub Copilot subscription model results in an average monthly loss of over $20 per user, with some heavy users losing up to $80 per month.

The third stage is the mandatory switch to pay by volume billing. On June 1, 2026, Microsoft announced that all GitHub Copilot plans will officially shift to token based billing, with a monthly fee of $19 directly converted into an equivalent token quota. This change will bring to the forefront the real costs that have long been hidden by the subscription system – according to Reddit community users’ calculations, one intelligent agent programming session can cost $30 to $40, and the monthly plan will be exhausted in a single use.

02
Bill Out of Control: When Tokens are More Expensive than People
The implementation of token based billing presents the true face of enterprise AI spending.

The billing figures on the enterprise side are shocking. Uber’s Chief Operating Officer Andrew Macdonald publicly stated in May 2026 that there is “no such thing as a line” between the growth in token consumption and the substantial improvement of the product, and coined a term specifically for this: “tokenmaxxing” to describe employees performing worthless tasks to boost usage.

A more direct data is that Uber exhausted its annual token budget in just the first four months of 2026; Salesforce expects to pay Anthropic approximately $300 million annually.

Anthropic’s own developer documentation shows that the average cost for developers using Claude Code is about $13 per workday, with 90% of users spending less than $30 per day – which translates to a 10 person development team spending over $75600 per year on tokens alone.

The input-output ratio is equally alarming. After summarizing data from 2444 companies, the enterprise data platform Entelligence.AI found that for every $1 invested in AI Token fees, only 18 cents generated actual value in reaching users; 44 cents are used to fix bugs introduced by AI itself, 27 cents go to rework, and 11 cents are consumed for review friction.

Faced with out of control bills, the enterprise side has begun to proactively control them. Amazon has suspended its internal AI usage ranking, requiring employees not to use AI for the sake of using it; Microsoft plans to gradually phase out Claude Code subscriptions for some key product department employees. Goldman Sachs pointed out that some companies have already spent 10% of their total labor costs on AI tokens, and this proportion may further increase in the coming quarters. This is not the disappearance of demand, but the end of the extensive era of AI spending.

03
Act 4: Price War Starts, OpenAI Considers Significant Price Reduction
It is in this context that the spark for the price war has been ignited.

According to The Wall Street Journal, Altman’s price reduction considerations were directly triggered by pressure to catch up with Anthropic. Anthropic’s revenue has recently grown significantly, and its programming tool Claude Code has become popular among software engineers. This five-year-old startup has even surpassed OpenAI’s valuation for the first time.

However, the cost of this price war will be exceptionally heavy. If the price is significantly reduced, it will further compress the already negative profit margins of the two companies, and the space provided by the competitive landscape is extremely limited.

The underlying risk that investors have long identified is that OpenAI and Anthropic’s products are highly substitutable, allowing customers to easily switch from one company to another. This means that even if prices are lowered to retain customers in the short term, they cannot truly build a moat, but only delay the loss of market share.

This dilemma is also transmitted through the financial cycle between cloud computing giants and AI labs.

According to corporate disclosure documents compiled by The Information, OpenAI and Anthropic together account for over half of Microsoft, Oracle, Google, and Amazon’s approximately $2 trillion commitment to future cloud services. If the price reduction triggers a downward revision in income expectations, this transmission chain will be under pressure in both directions.

American neuroscience and artificial intelligence expert Gary Marcus said, ‘This further exposes OpenAI’s vulnerability and demonstrates how serious the challenges it faces are.’. Once OpenAI goes downhill, it is likely to bring down companies such as Nvidia, Oracle, Coreweave, etc. The situation is rapidly deteriorating. ”

The long short divergence is openly confronting Wall Street. Morgan Stanley TMT analyst Mark Schilsky believes that the current bill anxiety is just the “minimum deceleration zone towards higher spending”: if the average price per million tokens decreases, but the penetration rate of AI payment by US companies continues to rise, the overall token usage will inevitably increase significantly mathematically; In addition, agentic AI has pushed the consumption of single task tokens to several times that of traditional question answering mode, and the long-term total expenditure is expected to be significantly higher than the current level.

Goldman Sachs semiconductor analyst Jim Covello takes a more pessimistic stance, believing that the current boom in the industry chain is directing almost all value towards semiconductor companies, a phenomenon that is “unprecedented and unsustainable in history”. Once companies face the real price of pay as you go, the capital flow that supports GPU procurement and model training will face a reversal.

04
Act 5: The Next Story of Token Economics?
After the price war, the next chapter of commercialization in the AI industry has not yet been written, but the outline is emerging.

Citadel Securities’ report provides a directional framework: tiered charging and pricing based on scarcity. The core logic is that inference intensive cutting-edge AI will not disappear, but will increasingly be concentrated in the hands of a few large enterprises that have the ability to bear the cost of computing power; For a wider range of enterprises, simpler models may be a more productive path until physical constraints are alleviated. This means that AI usage will move towards layering – high-value, complex tasks will continue to use cutting-edge models, while daily and batch tasks will shift towards cheaper or local models.

JPMorgan Chase holds a relatively optimistic view: even if the unit token price drops, the popularity of agent AI will double the token consumption of each task – existing data shows that the token consumption of each task can be reduced to 3.5 times the original after business agentization – the overall expenditure scale is still expected to continue to expand, and the current bill anxiety may only be the “minimum deceleration zone to higher expenditure”.

Nebius Chief Revenue Officer Marc Boroditsky proposed the concept of “valuemaxxing”, advocating for the industry to shift from pursuing maximum token consumption to truly generating value for each token. This direction is gradually becoming an industry consensus – but true commercial implementation still requires AI labs to find a pricing system that can reflect real costs and be accepted by enterprise customers, which is the core proposition of all current debates that have not yet been resolved.

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

According to June data from the US enterprise expenditure management platform Ramp, DeepSeek has topped the list of US enterprise software subscription growth rates. Ramp Chief Economist Ara Kharazian emphasized that this is not a local deployment of open source models, and that “enterprises are directly using DeepSeek to send and receive data”, which is a real paid direct connection use – he admitted that “he did not expect American companies to use DeepSeek”. According to third-party calculations, the average API price of DeepSeek V4 Pro is about one tenth of GPT-5.5 and about one eleventh of Claude Opus 4.7.

OpenAI and Anthropic are in a fierce competition, and the ultimate beneficiary may be the player who has already written “inclusive pricing” into their genes and does not need to explain profit margins to IPO investors. This may not be the most popular outcome of this price war, but it is becoming an increasingly difficult reality to ignore.