Even a powerhouse like Doubao will stoop for money.

Subscriptions, commission deduction, rate adjustment, mobile assistant development, deployment on intelligent cockpits... Doubao keeps moving forward and launches continuous attacks without a break.

Seeing through the essence behind the events, if we connect all these actions together, a main line gradually becomes clear: Doubao is frantically seeking revenue sources.

No matter who you are, even if you become the top player in the entire industry, you will eventually have to consider a core issue — monetization.

Doubao is no exception.

However, the embarrassing fact is that Doubao has hundreds of millions of users, with a truly huge user base, but its commercialization performance seems far from satisfactory. According to data disclosed by LatePost, as of the first half of the year, Doubao app generates less than 1 million yuan in daily revenue, which mainly comes from e-commerce commissions.

Not long ago, news spread in the market that Doubao's conversation team was laying off employees, which was interpreted as Doubao being unable to support itself, and ByteDance no longer wanted to continuously inject funds into it, so it gradually scaled down the team. Later, Liu Xing, the head of public relations of Doubao, refuted this rumor.

Speaking from the case of Doubao, this is not a problem unique to Doubao, but a common difficulty for all large model vendors. The only difference is that Doubao is backed by ByteDance, which has sufficient financial strength to support it for a period of time. For second-tier vendors, the challenges they face will be even greater.

Doubao is frantically seeking cash flow

Over the past two decades, Chinese internet companies have hardly come up with many new ideas in terms of monetization. Advertising, e-commerce, and value-added services are basically the only three options they have never broken away from.

What is even more regrettable is that after entering the fierce competition of AI large models, the commercialization methods of these large manufacturers are still the "old three", which is somewhat unenterprising.

Let's take a look at Doubao.

Since 2025, Doubao has launched a large number of intensive actions covering personnel rotation, organizational structure adjustment, product release and many other aspects. Rough statistics show that there are 40 actions in total, and new moves are launched almost every month.

Focusing on the commercialization level, the latest action is rate adjustment. In August 2026, multiple media reported that for hotel orders completed by redirecting from Doubao to Douyin Local Merchant platform, Doubao will charge separately, with a comprehensive rate of about 12%.

This is a microcosm of Doubao's efforts to seek revenue. Doubao's independent charging means that it will further strengthen its positioning as an entry portal. In fact, diverting traffic through conversations and finally deducting commissions from transactions has indeed become the exploration direction of Doubao's commercialization.

Doubao itself is very suitable for this role. According to data from QuestMobile, in September 2026, the monthly active users of Doubao's AI native App reached 419 million, a year-on-year increase of 143.6%. It has become a fact that users are increasingly inclined to ask questions through Doubao.

Conversations are scenarios, and where there are scenarios, there is the possibility of commercial conversion.

By the way, traditional advertising is paid promotion, where merchants use the "magic" of money to influence platform recommendations. But AI conversations are different. Large models deduct commissions only after the order is completed, which is equivalent to charging channel service fees.

Further back, the commercialization action that once sparked widespread discussion about Doubao was subscription charging. In June this year, while retaining the free basic version, Doubao launched paid subscription packages with three tiers of 68 yuan, 200 yuan and 500 yuan per month.

Doubao's intention is very obvious: with the continuous expansion of user scale and the rapid rise of computing power consumption, it wants to identify different user groups through subscription charging. The basic version mainly meets daily Q&A demands, with controllable computing power consumption, low cost and the ability to retain users. The paid version acts as a productivity tool, which can not only complete complex tasks, but also offset the increasingly rising cost curve.

Doubao has made a very clear plan, and this path is correct. But one basic premise should not be ignored: are C-end users willing to pay?

From past experience, after internet companies launch new products, users are generally accustomed to "getting free benefits" first, and will pay only when they find the product is valuable. After using large models, this situation goes even further: to "avoid" paying, users will use multiple products at the same time, switching back and forth between different large models.

At least judging from the user behavior around us, there are almost no users who have long-term and continuous subscriptions to Doubao at present.

The B-end is a major source of revenue for Doubao. Doubao outputs its model capabilities to the outside world through Volcano Engine, and enterprise customers pay according to the volume of model calls. Compared with C-end users, enterprise customers are generally willing to pay certain costs, which provides cash flow for Doubao.

In outdoor scenarios such as airports, massive large-screen advertisements put up by Doubao can be seen everywhere. Some media estimated that Doubao's annual advertising and marketing expenses in 2025 were as high as 15 billion yuan. Yiou once verified with relevant personnel from ByteDance, who said that this data was not accurate and there might be deviations in the statistical caliber.

But it is undeniable that Doubao's advertising volume is indeed very large. ByteDance cannot continuously "transfuse blood" for Doubao, and Doubao must have the ability to generate revenue on its own.

Furthermore, from a business perspective, Yiou believes that if a product cannot realize commercialization and no one is willing to pay for it, it can hardly be called a success.

Whether it is subscription packages, rate adjustments, mobile assistants or intelligent cockpits, Doubao has been constantly making attempts, whose essence is to explore revenue-generating paths through various methods.

Monetization Dilemma: Large Models Hit the Wall

"It's too difficult to monetize at the C-end!" More than one industry practitioner complained to Yiou.

From past internet products to today's large models, the problems they face are actually very similar — users are not willing to pay easily.

This is only a superficial problem. Traditional internet has economies of scale: once the product is launched, no matter how many new users are added, the marginal cost will be extremely low. In most cases, the more users there are, the more the cost will be spread thinner.

But large models have completely changed this situation. Every time a user performs a task such as asking a question, generating an image, or making a PPT, computing power will be called. The more complex the task is, the more Tokens will be consumed.

The problem lies here: as Doubao's monthly active users grow from millions to tens of millions, then to hundreds of millions, the more users there are, the higher the computing power cost will be, and the bill will continue to rise.

For a long time in the past, the internet industry generally believed that having users means having traffic. Large models have not changed this consensus either. In the early stage of competition, all players would compete for user scale. This is just like marching to fight: troops are the core assets.

But the change is that large models are in a conversational and execution mode, which makes the value monetization process more complex. If effective monetization cannot be achieved, the huge number of users will instead increase the burden, forming a "diseconomies of scale" effect. In other words, if the troops cannot fight, they will consume more food and grass instead.

What's more difficult is that the traditional internet advertising strategy is failing on large models. As mentioned earlier, the traditional approach is to first expand users and traffic, and then monetize through advertising. Domestic large manufacturers such as ByteDance, Tencent and Baidu all follow this model without exception.

But when this mature strategy is applied to Doubao, it will face a contradiction.

Why do users use Doubao? Most users have clear demands, whether it is simple Q&A or executing task instructions. Users hope to operate continuously, and no one wants a sudden advertisement to pop up when using a large model.

You can insert advertisements in short videos, but you cannot let advertisements pop up suddenly when users are using Doubao. Interrupting users' instructions will greatly affect their user experience. So it is very simple: large models themselves are not suitable for inserting advertisements on a large scale.

In August 2026, the half-year report disclosed by MiniMax shows that open platforms and other AI-based enterprise services have become the largest business, accounting for 63.4% of total revenue. Back in 2024, this business only accounted for 32.8% of total revenue, while AI native products accounted for as high as 71.4% of revenue at that time. Obviously, the revenue pillar has been replaced by the B-end. As one of the only two listed large model companies in China, MiniMax can be regarded as a typical sample of the industry.

Although we cannot generalize from this case, we can at least judge that it is very difficult to make money solely relying on C-end users, especially for ordinary users who only use the product for conversations, searches and simple tasks, they can hardly become the main revenue source at the current stage.

In a word, Doubao's urgent pursuit of commercialization is, to a certain extent, the result of cost driven by the surge in users. Douyin's main business cannot continue to provide subsidies, and monetization is a rigid requirement.

After Entering the Value Realization Stage, the Industry Starts the Knockout Round

Large models seem to have technical barriers, but in fact, the competitiveness of each company's products in the C-end market is uneven.

For example, when Doubao launched paid subscriptions, it once sparked controversy, which also shows that it seems not to have reached a unanimous consensus in the market. Qwen, Yuanbao, Kimi, DeepSeek and other large models are facing the same problem. Second-tier large models such as Zhipu Qingyan, Baixiaoying, Wanzhi, Tiangong AI and iFlytek Spark may face even more difficult situations.

Without obvious differentiated advantages and low user stickiness, users do not have to choose your product, so it is difficult for manufacturers to raise prices. Unless they have irreplicable capabilities, once they raise the charging standard, users will immediately turn to other products. The so-called tens of millions or hundreds of millions of users look more like "castles in the air".

The commercialization pressure faced by Doubao is like a mirror placed in front of the entire large model industry. In the expansion cycle, large model manufacturers are used to following the mobile internet strategy, taking user scale as an important indicator of competition.

There may have been some illusions before that AI products can replicate the advertising model to achieve monetization through conversations. But Doubao's exploration shows that the underlying logic of task-based and continuous conversational AI products is contradictory to advertising. At least for now, this road has not been fully opened.

In this case, the evaluation criteria of the entire track will also be rewritten. In the past, competing for monthly active users could reflect the strength of a company, but now this indicator is almost invalid, so in the future, we need to consider how to calculate the cost bill.

Offering free services will lead to losses, which requires continuous subsidies to survive; while raising the price too high will shake the user base. Doubao is not an isolated case, all players in the track are facing the same dilemma.

The difference is that Doubao still has ByteDance's huge ecosystem as support, while second-tier vendors and startup large model companies lack both ecological test scenarios and continuous capital injection, so they will face even more difficult situations.

Since ChatGPT was launched at the end of 2022, the global large model industry has set off a financing boom, with capital pouring in, almost replicating the grand occasion of the mobile internet era, when a company could easily obtain hundreds of millions of yuan in financing just with a PPT. It is hard to imagine that this can happen in today's investment environment.

But no wave keeps rising, the economy operates in cycles, and so does industrial development. After years of vigorous development, large models have evolved from conversations to task execution, from general capabilities to applications in vertical scenarios across thousands of industries. Large models have gradually moved from the money-burning expansion stage of competing for parameters and attracting users, to the value realization stage. A round of knockout competition will be staged soon.

AI is not a technology in a certain field, but a fundamental industrial revolution, which determines that products must be inclusive and cannot become a game for only a few people.

Doubao is a sample. From this large model product with the largest user scale in China, we can see the division and pain that artificial intelligence experiences in different development stages.

This article is from the WeChat official account "Yiou" (ID: i-yiou), written by Wang Yuanlei, and published with authorization from 36Kr.