How long can the big model token business last?
Where will the future of the large-scale token billing model go? This article deeply explores the current situation, challenges and transformation direction of the token economy.
Core content:
1. The current situation of traffic dividends and price involution of the token billing model
2. Challenges faced by the business model such as the single economic model and the surge in data costs
3. Three possible directions for the industry to transform from "selling tokens" to "selling services"
Preface
The business model of charging by token for large models has become the mainstream of the industry since its birth, but how long can the vitality of this model last? From the "floor price" strategy of Volcano Engine to Huawei's prediction of a 33-fold increase in token traffic, from Zhou Hongyi's assertion that "the dividend period is at least 10 years" to Scale AI's annual revenue of US$1 billion through data services, the industry is undergoing a deep transformation from quantitative change to qualitative change. This article will briefly discuss the possible future direction of the Token economy from the perspectives of technology, business and ecology.
The current status of the token economy: from traffic dividends to price involution
At present, the core contradiction of the Token billing model lies in the game between scale expansion and diminishing marginal returns . Volcano Engine has reduced the unit price of Token to 0.0008 yuan/thousand Tokens through engineering optimization, driving the daily average call volume from 120 billion to 500 billion, confirming the logic of "lowering prices for market share". Huawei data shows that the Token traffic in the Chinese market has increased 33 times in 8 months, but the price per Token has dropped 97%. This trend of "increasing volume and falling prices" is similar to the early development path of cloud computing, and is essentially a necessary stage for the popularization of technology.
However, the limit of price war has been reached. When the unit price of token approaches the marginal cost, the stimulating effect of price reduction on the market will be significantly weakened. Tan Dai said frankly: "Further price reduction is of little significance, and the improvement of model capabilities is the key." This indicates that industry competition will shift from the price dimension to the value dimension , and enterprises need to create higher technical added value within the unit token.
Business model challenges: the ceiling of single billing
The limitations of the Token model are gradually being exposed:
Single economic model : Current billing only reflects computing power consumption, not the actual value of the model output. For example, the cost of generating 100 tokens of code is the same as the cost of 100 tokens of chat content, but the former may be several times more valuable to users.
Data costs surge : The Scaling Law of large models requires exponential growth in data volume, but the cost of high-quality data annotation remains high. Scale AI earns $1 billion a year by building a "data factory", which just shows that data services have become a new track independent of Tokens.
Ecosystem dependence risk : Over-reliance on leading customers (such as Meta and Google) may lead to an imbalance in the revenue structure. This issue has caused investor concerns in companies such as Scale AI.
These challenges have forced companies to explore diversified business models. Volcano Engine has proposed that it may shift to "pay by problem-solving effect" in the future, while Huawei has deployed AI-Centric network architecture, trying to extend the value chain by optimizing traffic carrying capacity.
Future transformation direction: from "selling tokens" to "selling services"
The breakthroughs in industry transformation may focus on three directions:
Capability-tiered pricing : Gradient pricing is designed based on the complexity of model output (such as code generation, multi-round dialogue, and long text understanding) to deeply bind billing with value. For example, OpenAI's o1 series has subdivided the pricing standards for scenarios such as programming and instruction following through the SEAL ranking.
End-to-end solution : Encapsulate model capabilities into vertical scenario solutions (such as legal document generation, medical diagnosis assistance), and use subscription or performance sharing instead of token billing. The Jomoo Group's access to DeepSeek to achieve intelligent transformation of the bathroom industry is a typical example of this model.
Ecosystem collaborative innovation : Through open APIs, developer tool chains and computing power scheduling platforms, we build an ecosystem with Token as the traffic entrance. After Volcano Engine attracts customers through low-price strategies, it promotes them to restructure their IT architecture and purchase more cloud services, which is the embodiment of this logic.
The underlying logic supporting long-term development
The vitality of the Token economy ultimately depends on the resonance between technological breakthroughs and demand evolution :
Technical side : Improvements in capabilities such as hallucination control and long-context understanding will expand the boundaries of model application. For example, the Doubao large model optimized the engineering architecture to increase the call volume by 33 times while keeping costs under control.
Demand side : Generative AI is shifting from an auxiliary tool to the core of productivity. Huawei predicts that by 2030, the average daily network traffic driven by AI will reach 500TB, more than five times the current mobile traffic, which means that there is still huge room for growth in token demand.
Data side : Scale AI has demonstrated the commercial value of high-quality data services through the "human expert + AI" data annotation model. The future Token economy may need to be deeply integrated with data services to form a closed loop of "data-model-application".
In conclusion: Token will not disappear, but must evolve
As the "measurement" of the big model era, Token will remain the mainstream billing unit in the short term, but its connotation will undergo fundamental changes. The future Token economy may present the characteristics of "free basic resources and monetization of value-added services" - just like cloud computing has shifted from charging by server rental to charging by API calls. When the model capability is strong enough, Token may degenerate into an underlying billing unit, and the real profit will come from the intelligent service ecosystem built on Token. The end point of this change may be an era of "Token invisibility": users no longer perceive the existence of Tokens, and only pay for the actual problems solved by AI.
