Years from now, when people reflect on China’s AI application market in 2026, Doubao’s move to start charging may well be remembered as a pivotal turning point.

Certainly, Doubao is not the first AI product to introduce fees, nor is it the first internet product to attempt to convert free users into paying ones. However, its uniqueness stems from the fact that this shift is occurring on an AI platform with hundreds of millions of monthly and daily active users, and it is being spearheaded by ByteDance, a company renowned for its ability to monetize traffic.

Thus, while Doubao’s decision to charge may appear on the surface to be just another app introducing a membership program, it actually signifies ByteDance beginning to take a more pragmatic approach to its AI strategy.

According to a report by Cailian Press, Doubao is expected to roll out paid content in late June and update related functions at the same time at the Force conference. Previously, a paid version service declaration had appeared on Doubao’s App Store page, indicating three subscription tiers: Standard, Enhanced, and Professional, with monthly subscription prices of 68 yuan, 200 yuan, and 500 yuan, respectively.

This development has yet to be officially confirmed by ByteDance. According to a previous report by Yicai, Doubao’s official response was cautious: “Doubao has always provided free services, and on top of these free services, it is also exploring the introduction of more value-added services. The details of relevant plans are still in the testing phase.” This statement warrants careful consideration.

It implies that Doubao is not planning to put all users behind a paywall. Ordinary Q&A, daily chats, and basic usage will likely remain free. What may enter the paid system are functions that are more complex, computationally intensive, and closer to productivity tools.

According to sources close to Doubao, the paid functions will primarily focus on complex tasks and productivity scenarios, such as PPT generation, data analysis, and film and television production.

This is the true essence of Doubao’s charging model: not abolishing free services but establishing a pricing system on top of them.

Over the past two years, China’s AI application market has mirrored the early days of mobile internet expansion. Models need to be fast, products lightweight, entry points numerous, and user growth rapid. The more free a service is, the more likely it is to be tried by users; the larger the monthly active user base, the easier it is to attract capital and market attention.

Doubao has emerged as the most typical winner in this round of competition.

QuestMobile data shows that as of March 2026, the monthly active user base of AI-native apps has reached 440 million. Among them, Doubao, Qianwen, and DeepSeek rank in the top three, with monthly active users of 345 million, 166 million, and 127 million, respectively. Doubao’s user base alone exceeds the combined total of Qianwen and DeepSeek.

What does having over 300 million monthly active users mean in the history of China’s internet?

It means that a product is no longer just a tool but has the opportunity to become an entry point. It also means that user habits, usage scenarios, and commercialization potential have reached another level.

However, AI entry points differ fundamentally from previous-generation internet entry points: they are expensive.

When a user watches short videos, the platform primarily bears costs related to bandwidth, recommendation, moderation, and operations. When a user asks an AI question, it involves model inference, computational resource allocation, server resources, engineering maintenance, and continuous training. The traditional internet logic is that the larger the scale, the lower the marginal cost. The challenge with AI applications is that the more active users are, the clearer the bill becomes.

Volcano Engine previously disclosed that as of March 2026, the daily average token usage of Doubao’s large model had exceeded 120 trillion, compared to approximately 120 billion when the model was first released in May 2024. This represents a roughly 1,000-fold increase over two years.

Of course, these 120 trillion tokens cannot be simply equated with the consumption by Doubao’s consumer-facing app users. They also include various usage scenarios such as enterprise clients, ByteDance’s internal operations, and the developer ecosystem. However, this figure is sufficient to illustrate a fact: ByteDance’s AI capabilities have entered a stage of high-frequency, resource-intensive, and large-scale operation.

When an AI product transitions from a novel toy to a daily tool, free services cease to be just a customer acquisition strategy and become a continuous computational cost.

This is why Doubao’s move to charge is not sudden. It is not a simple “membership trial” but a signal of ByteDance’s AI business shifting from a growth logic to a unit economic model.

The so-called unit economic model, simply put, means determining whether serving an additional user, completing an additional call, or generating an additional piece of content results in profit or loss.

In the past, ByteDance excelled at the traffic flywheel. Today’s Headlines, Douyin, and later Douyin E-commerce essentially improved content distribution efficiency through algorithms and then monetized through advertising, e-commerce, and transactions. The longer users stayed, the more content was supplied, and the larger the commercialization space became.

However, AI is not simply about content distribution. AI answers questions, generates content, processes documents, handles images and videos, and will eventually perform tasks on behalf of users. It does not just push existing content more efficiently to users but consumes computational resources in real-time to produce new results for users.

This determines that AI commercialization cannot rely solely on free services to achieve scale.

Doubao must answer a more fundamental question: With so many users and calls, who will ultimately foot the bill?

If we roughly calculate based on Doubao’s 345 million monthly active users, even if only 1% of users pay 68 yuan per month, the monthly revenue would be approximately over 200 million yuan. If the payment rate reaches 5%, the annual revenue would be approximately in the tens of billions of yuan. While this figure is not small for an app, when placed in the context of ByteDance’s AI infrastructure, model training, chip procurement, and data center construction, it still resembles more of a buffer than the final answer.

Therefore, what Doubao truly aims to validate is not just whether users are willing to spend 68 yuan on AI.

It aims to validate whether a Chinese domestic AI assistant can transform from a free tool into a sustainable business system.

This is not an easy task.

Chinese users are not entirely unaccustomed to paying. They pay for video streaming sites, cloud storage upgrades, office software, and in-game items. However, AI subscriptions face an awkward situation: their value has not yet been stably perceived by users.

Video streaming memberships sell the certainty of content, cloud storage memberships sell the certainty of space, and office software sells the certainty of functions. What AI assistants sell is more abstract: it could be an answer, a document, a PPT, a search summary, an image, or even an automated task execution in the future.

If it's just “faster answers,” “more calls,” or “a stronger model,” many users find it difficult to continue paying.

Because there are too many alternatives.

Today, users can use Doubao, tomorrow they can use Yuanbao, and the next day they can use Qianwen or DeepSeek. AI assistants lack the strong relational chains of WeChat or the deep content immersion and algorithmic inertia of Douyin. As long as the functional differences are not significant, the cost of user migration is low.

This is also Doubao’s biggest challenge: it must not only make users feel that it is good to use but also make them feel that it is indispensable.

What truly has the potential to charge is not casual chatting but results.

A directly deliverable PPT, a clearly structured report, a publishable video, a reusable data analysis, or a task that saves two hours of work—these are where AI can truly establish a price.

In other words, the success of Doubao’s membership does not depend on how beautifully the pricing table is designed but on whether it can show users a clear equation: I spent money, so I saved time, improved efficiency, or even gained income.

However, if we only focus on membership, we still underestimate ByteDance.

What deserves the most attention about ByteDance’s Doubao is not how much membership revenue it can generate but whether it can integrate AI Q&A, e-commerce recommendations, and transaction payments into a single chain.

Media reports mention that if the subscription model progresses smoothly, Doubao may further combine e-commerce functions to update paid scenarios in the third quarter and drive traffic to Douyin Mall through subsidies. This is the most informative part of the entire matter.

Because it indicates that subscriptions are likely not the endpoint but merely the first gate.

In the past, Douyin E-commerce relied on content to drive purchases. A short video, a live stream, or an influencer would lead users from interest to purchase. In the future, Doubao may provide another path: users first express their needs, AI helps compare products, explain differences, and provide suggestions, and finally directs the purchase behavior to Douyin Mall.

The former is content e-commerce, while the latter is more like decision-making e-commerce.

This is a significant change.

Content e-commerce addresses “I didn’t intend to buy, but after watching, I want to,” while decision-making e-commerce addresses “I intended to buy, but I didn’t know what to buy.” The former relies on stimulation, while the latter relies on trust. The former depends on scenarios and emotions, while the latter depends on understanding and judgment.

If Doubao can succeed in this area, it will not just be a chatting app but could become ByteDance’s new search box, new shopping guide entry point, and new productivity tool in the AI era.

This is ByteDance’s true calculation.

It does not necessarily have to replicate ChatGPT Plus or rely solely on subscriptions for revenue. A path more in line with ByteDance’s DNA would involve maintaining scale through free entry points, screening high-value users through membership subscriptions, undertaking B-end calls through Volcano Engine, completing transaction conversions through Douyin E-commerce and advertising, and deepening task execution capabilities through intelligent agent services.

This is a hybrid commercialization path.

And this path is becoming a common choice for AI companies globally.

Even ChatGPT has not solely bet on subscriptions. OpenAI officially describes its business model as a multi-tiered structure: consumer subscriptions, team subscriptions, an ad-supported and commercially supported free tier, and usage-based API billing. In other words, once AI applications reach a sufficient scale, a single subscription model is unlikely to support all costs. Advertising, e-commerce, APIs, enterprise services, and outcome-based payments will all become part of the commercialization puzzle.

The domestic market has already provided a reminder. Baidu’s Wenxin Yiyan once launched a membership service but later announced it would be free again and processed membership refunds. This indicates that relying solely on a “stronger model” to sell C-end subscriptions is not easy in the Chinese market.

Users are not unwilling to pay for AI but are unwilling to pay for an AI that can be easily replaced at any time.

Therefore, Doubao must take a different path. It holds three cards. The first card is scale. With 345 million monthly active users and hundreds of millions of daily active users, Doubao possesses the strongest real-world usage scenarios among domestic AI applications. The more users there are, the more questions there are, the denser the feedback, and the easier it is to refine model and agent capabilities in response to real needs. The second card is the ecosystem. Doubao is not an isolated product but is backed by Douyin, e-commerce, advertising, search, Feishu, Volcano Engine, and Kouzi.

It can integrate AI capabilities into content, transactions, office work, and enterprise services rather than being confined to a chatting window.

The third card is organizational and engineering capabilities. The competition among large models is not just about parameters and rankings but also about infrastructure, inference costs, product iteration, and engineering efficiency. The recommendation systems, traffic operations, and commercialization capabilities that ByteDance has accumulated over the years still hold transferable value in the AI era.

However, the risks are also clear.

The first risk is that users may find it expensive.

Here, “expensive” does not refer to the 68 yuan figure itself but to users not understanding why they should pay this 68 yuan. If the paid version merely offers more functions and a stronger model without creating a clear sense of outcome, it can easily be diluted by free competitors.

The second risk is trust.

AI assistants differ from search engines. Search engines provide users with a list of links, allowing them to judge for themselves; AI assistants provide direct answers. Because they provide direct answers, users have higher expectations for their objectivity.

If Doubao incorporates product recommendations, advertising, and e-commerce referrals into its answer system in the future, it must carefully manage the boundary between commercial recommendations and objective advice. Once users feel that answers are driven by sales, trust will be eroded.

For AI products, trust is the moat.

The third risk is the return cycle.

The AI infrastructure demands substantial asset investments. The training of models, the procurement of chips, the construction of data centers, the maintenance of inference calls, and the attraction of top talent all necessitate long-term financial commitment. Even with growth in membership revenue, it is improbable that all costs will be promptly covered. The true determinant of Doubao's long-term value lies not in the initial number of users who sign up for membership, but rather in its ability to translate high-frequency usage into steady revenue streams, transaction conversions, and ecosystem dominance.

Therefore, Doubao's decision to implement charges should not be viewed as a standalone product update.

Instead, it represents a pivotal test case for the commercialization of AI applications within China.

Over the past two years, AI companies have been vying to demonstrate superior model capabilities, faster product development cycles, larger user bases, and a greater propensity for offering complimentary services. Moving forward, the industry is poised to transition into a new phase of competition: the race to achieve lower inference costs, unlock more high-value use cases, and construct a viable commercial closed loop—all while preserving user trust.

Doubao has already attained a significant scale.

However, scale alone does not constitute a business model. Rather, it serves as a prerequisite for entering the commercialization arena.

Going forward, Doubao must demonstrate that an AI application with hundreds of millions of users can evolve from a free entry point into ByteDance's new search engine, new shopping guide, and new productivity platform in the AI era.

Doubao's move to charge users is not merely about initiating fees; it is about establishing the inaugural pricing framework for ByteDance's AI super entry point.

What truly warrants attention is not the number of users who sign up for membership in late June, but rather whether ByteDance can transform AI—a costly traffic asset—into a sustainable and profitable business venture.