Can Doubao Spark a Wave of AI Subscriptions in China?
Author | Wang Xiaojuan
The domestic AI native application market has reached an inflection point of industry significance.
On May 4, ByteDance's AI app Doubao updated its paid version service statement on Apple's App Store, officially announcing the launch of a three-tier subscription model—Standard (￥68/month or ￥688/year), Enhanced (￥200/month or ￥2048/year), and Professional (￥500/month or ￥5088/year)—while keeping the basic free version.
This move not only marks Doubao, the AI application with the largest user base in China, as the first to take substantial steps toward C-end commercialization, but also signals the end of the phase where domestic large model applications rely solely on free scaling.
According to the latest QuestMobile Spring 2026 report, as of Q1 2026, domestic AI native app monthly active users have reached 440 million.
Among them, Doubao dominates with 345 million monthly active users, exceeding the sum of the industry’s second, Qianwen (166 million), and third, DeepSeek (127 million).
However, massive user numbers don't simply translate into scale effects, but rather exponential computing power consumption and sustained pressure on profits. Doubao's paid trial aims to alleviate the cost pressure of high computing scenarios by charging core productivity users, establishing a sustainable commercial cycle.
As the leading app to break the “all free” industry tacit understanding among major players, Doubao's paid attempt has notable industry significance. Morgan Stanley's latest research report defines Doubao’s switch to paid mode as a key node from the user education phase to commercialization in the industry.
Yet, facing Doubao's paid model, some users are sure to flow to other free platforms. But does this mean competitors like Qianwen and Yuanbao will reap the traffic and stick to the free model long-term? In the face of computing costs, is this traffic migration a benefit or a burden for recipients?
These questions will be answered only after Doubao's paid model officially launches.
The "Sweet Burden" of 345 Million Monthly Actives
Looking back at Doubao’s growth curve, it’s a trajectory envied by any internet product.
Throughout 2025, Doubao's MAUs soared from 99.8 million in Q1 to 227 million in Q4, a 127% increase, becoming China's first AI native app to surpass 100 million daily actives. In 2026, growth continued; in Q1 alone, it gained about 100 million new actives, with Spring Festival Gala DAUs reaching 145 million.
More noteworthy is growth quality. According to AppGrowing, a domestic mobile ad monitoring agency, Doubao’s spending declined in 2025: ￥161 million in Q1, ￥117 million in Q2, halving to ￥65 million in Q3, then slightly rising to ￥92 million in Q4. Meanwhile, Doubao's 30-day retention rate averaged 44% from Jan-Nov 2025, far ahead of industry No. 2.
It’s this “low acquisition, high retention” growth model that puts Doubao miles ahead of peers, making it No. 1 in domestic AI applications.
However, the underlying economics of large model apps differ fundamentally from traditional mobile internet products.
Traditional SaaS or social software have near-zero marginal cost; adding one user costs almost nothing. But for generative AI, every interaction and inference really consumes GPU cluster computing power.
According to a computing cost breakdown from CNDS, hardware depreciation accounts for 58% and power consumption for about 29% of the cost of a single Doubao inference. This means the 345 million MAUs’ 120 trillion daily tokens translate directly into millions per day in power and hardware loss bills.
From a profit perspective, “big DAU equals big debt” is clearly unsustainable in the long run.
With model upgrades, especially the advent of multimodal and long-text parsing, single task inference costs are rising geometrically. Generating a professional PPT or rendering a few minutes of video might consume tens or even hundreds of times the computing power of ordinary chat. If all users continue to receive undifferentiated free services, the platform’s financial model may be dragged under by a minority of heavy-demand users.
Many Doubao users told Wallstreetcn that during recent experience, complex tasks like deep research and video generation often faced network queues during peak hours, reflecting Doubao's tightness in computing power for free services.
Under such pressure, “free” is no longer sustainable. People close to ByteDance said its net profit fell more than 70% YoY in 2025, with massive AI capital expenditures a major drag.
Going forward, Doubao will adopt permanent free basic + paid value-added services in a layered architecture. Sources close to Doubao revealed the paid features will focus mainly on complex tasks and productivity scenarios such as PPT generation, data analysis, video production, etc.
In other words, Doubao uses three pricing tiers to precisely identify heavy computing power users among its 345 million MAUs.
Looking at the pricing strategy, Doubao’s tiers are ￥68, ￥200, and ￥500 per month, with the entry price almost 40% higher than domestic competitor Kimi’s ￥49/month. Morgan Stanley points out the pricing targets professional users rather than the mass market, aiming at creators and knowledge workers.
This means Doubao's core monetization logic is not converting free users to paying ones, but screening high-value users with strong willingness and ability to pay.
Morgan Stanley’s analysis calculates that, even with a conservative 0.3% to 3% paid conversion rate, Doubao’s annualized subscription revenue could reach between $101 million and $1.5 billion; in a neutral scenario, about $426 million to $684 million. Compared to ByteDance’s core ad business, this is limited in scale but enough to open the first window for user monetization.
In fact, Doubao’s commercialization is not sudden.
ByteDance’s AI development platform “Kouzi” has been charging professional developers since 2024, launching a personal version in late January 2026, priced from ￥19.9 to ￥99/month. Doubao charges may just be a natural extension in ByteDance's systematic AI matrix monetization.
Game Across the Paywall
Once Doubao's paid statement came out, market attention quickly shifted to competitors’ reactions.
With Doubao’s paywall established, many users began worrying about limitations on free features, spurring inevitable discussion of user loss in the industry.
For example, some geek or heavy productivity users used to getting advanced features for free may choose to migrate to currently free Qianwen or Yuanbao when faced with monthly fees of ￥68-￥500.
From short-term data, this overflow effect is inevitable.
Many users are already seeking alternatives. A professional skills training user told Wallstreetcn she often used Doubao to generate PPTs for class, saving much time, but would seek alternatives if she had to pay at least ￥68/month. However, due to habit, she decided to batch generate PPTs before the paid version goes online.
However, given the large model industry’s cost paradigm, user migration could also become a "sweet burden" for recipients—after all, complex tasks mean high inference costs.
If Qianwen or Yuanbao absorb these heavy users leaving Doubao and continue to provide free long text parsing, deep data mining, and video generation indiscriminately, their server and electricity expenses will quickly explode.
In today’s macro environment where major internet companies emphasize cost reduction and efficiency and self-sustaining operations, no company can bear the computing power burden squeezed out by competitors indefinitely. Doubao breaking the “free tacit agreement” may trigger a wave of paid AI applications across major internet companies.
An AI industry worker told Wallstreetcn that for complex scenarios, paid models are inevitable in the future to continuously provide high-quality service. He often pays to test domestic and international AI models for work, and as a senior user, he believes users do not just care about paid versus free; if the model is smart enough and markedly improves efficiency, many productive users are willing to pay.
From an industry evolution perspective, Doubao’s leading paid move may be the start of the domestic large model industry “testing the water”; as a nearly 400 million MAU national-level app, Doubao may break the previous "who charges first dies first" dilemma.
Short term, it’s unclear if Qianwen, Yuanbao, etc. will follow suit; but mid-to-long term, the costs of heavy computing power users will inevitably show up on financial reports, and these players are likely to follow Doubao, launching “basic free + advanced paid” scenarios for complex tasks.
Ultimately, free is just a temporary market education tool; equivalent value exchange is the final form of business.
In this process, Doubao’s commercialization path can be compared to industry pioneer ChatGPT’s paid path.
OpenAI also endured huge cost pains early on and then established a strict layered subscription system. Currently, ChatGPT has not only a basic free version, but also an $8/month Go version, $20/month Plus, Team, and a $200/month Pro edition.
In comparison, Doubao's pricing logic is highly similar to ChatGPT’s, but prices have been localized for the Chinese market.
Doubao’s Standard ￥68/month is lower than ChatGPT Plus, aiming to lower the payment threshold for domestic users and cultivate a subscription habit; while its Professional ￥500/month, though quite high, is still restrained compared to ChatGPT Pro’s $200/month.
ChatGPT’s financials have shown that as long as inference really improves productivity, C-end users have strong willingness to pay.
Doubao’s paid trial is essentially testing whether this logic holds in China.
For the industry, this marks an elevation of competition; the past two years saw fierce price wars in China’s large model B-end API market, compressing commercial value. Now, in the C-end, if Doubao successfully runs a high customer value subscription model, the industry will shift from scale-focused competition to a blend of capabilities and business models.
Back to its paid model, in the short term, this may shift some users to other free platforms and even trigger debates on pricing rationality; but in the long run, it pushes the AI industry back to rational pricing mechanisms.
The final test will still return to product capability—when products are strong enough to meaningfully boost productivity, users may pay for consumed computing power and R&D achievements, which may form the basic consensus for healthy AI industry development in the next stage.
Risk Disclosure and DisclaimerThe market carries risks; investment needs caution. This article does not constitute personal investment advice, nor takes into account individual users' special investment goals, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article suit their specific circumstances. Investments based on this are at your own risk.