710 Million Users, One Hard Question: Can China’s AI Apps Turn Scale Into Profit?

China's top four AI native apps—Doubao, Qianwen, DeepSeek, and Yuanbao—collectively logged 710 million monthly active users in May 2026, yet the industry's defining question has shifted from growth to survival economics: at what point does scale become a liability rather than an asset.

The aggregate figure, drawn from QuestMobile's May 2026 ranking of the top 10 AI native apps, obscures a fractured competitive landscape where each major player is executing a materially different monetization thesis. The data arrived the same week that DeepSeek confirmed a fundraising round exceeding RMB 50 billion (approximately US$6.94 billion) at a post-money valuation exceeding US$50 billion—a structural financing event that signals the AI infrastructure arms race is accelerating, not plateauing.

Market observers note the timing is not coincidental. Compute costs are compressing margins across the sector simultaneously, forcing a strategic divergence that will likely determine which business models survive into 2027.

Doubao's Free-User Model Cracks Under Compute Pressure

ByteDance's Doubao remains the undisputed leader with 368 million MAU in May 2026, posting year-over-year growth of 182%. The number, however, masks a structural stress test that is now playing out in real time.

On May 4, Doubao quietly listed tiered subscription plans on the App Store: RMB 68/month for standard access, RMB 200/month for enhanced, and RMB 500/month for professional. The immediate market response was instructive: third-party data shows approximately 6.1 million users churned within the month, and a survey of over 90,000 respondents found 43% stated they would discontinue use upon any paywall introduction.

The churn is manageable in absolute terms—QuestMobile data shows Doubao's MAU still expanded by roughly 20 million between March and May 2026. But the episode exposed the structural fragility of a user base built on three years of zero-cost access: switching costs are near zero, and user loyalty is shallow.

The compute economics driving the pricing test are unambiguous. Doubao's daily token call volume surpassed 1.2 quadrillion as of March 2026. Separately, ByteDance's 2025 net profit reportedly declined by more than 70% year-over-year, with AI compute infrastructure cited as a primary driver. ByteDance has not publicly confirmed the profit figure.

The three-tier pricing structure reads less as a revenue strategy than a segmentation exercise. By identifying price-insensitive heavy users now, ByteDance can build a paying cohort while retaining the broader free base as a data and engagement flywheel. The company's disclosed commercial roadmap routes monetization through Douyin e-commerce integration—the "conversation-to-transaction" pipeline—with material revenue contributions targeted for Q3 2026 and beyond. The RMB 68 subscription, in this framing, is a user education campaign, not a P&L line item.

Qianwen's 5,303% Growth Masks Post-Promotion Decay

Alibaba's Qianwen posted the most eye-catching headline number on the QuestMobile chart: year-over-year MAU growth of 5,303%, reaching 162 million in May 2026. The denominator effect is significant—Qianwen was rebranded from "Tongyi" in November 2025, resetting its baseline—but the growth is also heavily event-driven.

The inflection point was the 2026 Lunar New Year. On February 6, Alibaba launched its "Spring Festival Hosting Plan," a RMB 3 billion (approximately US$416.7 million) cross-platform promotion that allowed users to issue purchase commands to Qianwen, which would then execute and subsidize the orders. First-day order volume exceeded 10 million; daily active users spiked by 51 million in a single session; Qianwen briefly surpassed Doubao at the top of the App Store charts.

The critical question is retention. Comparing March to May 2026 data, Qianwen recorded the largest absolute MAU decline among the top four—a directional signal, even if the magnitude remains modest. Promotional cohorts acquired through subsidy programs historically exhibit structurally lower retention than organically acquired users.

Alibaba's structural moat, however, is real and distinct from the marketing spend. The Taobao fulfillment infrastructure, Alipay payment rails, and Alibaba Cloud compute stack enable Qianwen to execute a genuine end-to-end "say it, buy it, receive it" loop—a capability that pure-play AI startups cannot replicate on any near-term timeline. The question is whether that infrastructure advantage can sustain MAU without continuous subsidy. Current data suggests it cannot yet do so independently.

DeepSeek Raises RMB 50 Billion While Redefining What "User" Means

A 24% MAU decline for DeepSeek in May 2026 is analytically misleading. The company's C-end app serves as a public interface; its core business is API infrastructure for developers and enterprises. QuestMobile's March 2026 data showed DeepSeek's monthly usage frequency per remaining user at 41.7 sessions—still growing year-over-year—indicating that the departing cohort consists primarily of curiosity-driven users who arrived during the R1 model's viral moment in early 2025.

The April 2026 product and commercial moves are more consequential. DeepSeek open-sourced its V4 model: 1.6 trillion parameters, one-million-token context window, and—critically—a complete, reproducible training migration path from Nvidia CUDA to Huawei's Ascend chip stack. This marks the first publicly documented end-to-end Ascend training scheme for a frontier Chinese model, carrying engineering significance that extends well beyond benchmark performance. Simultaneously, V4-Pro API pricing was permanently reduced to 25% of its prior level, deepening the cost moat against developer alternatives.

The financing structure disclosed on June 16, 2026, via The Information represents a more profound shift. The round totals over RMB 50 billion (approximately US$6.94 billion) at a post-money valuation exceeding US$50 billion. Structurally, with one exception—a state-level fund—all external capital was directed not into DeepSeek directly but into a limited partnership managed personally by founder Liang Wenfeng. Investors receive no voting rights and face a five-year lock-up. Liang had previously locked his personal control stake at 84.29%.

Disclosed investors include Liang himself at RMB 20 billion (approximately US$2.78 billion), Tencent at RMB 10 billion (approximately US$1.39 billion), CATL at RMB 5 billion (approximately US$694 million), and JD.com and IDG Capital at RMB 3 billion (approximately US$416.7 million) each. Alibaba, which had been reported in April as a prospective investor, did not appear in the final roster.The capital structure is a deliberate governance statement: external money is welcome; external influence is not. DeepSeek is transitioning from a self-funded research lab into a state-industrial capital hybrid—but one where founder control remains effectively absolute.

Tencent Hedges Yuanbao Weakness With a Dual-Track Bet

Yuanbao, Tencent's standalone AI app, is the most transparent underperformer in the top four. A RMB 1 billion (approximately US$138.9 million) Spring Festival social-sharing campaign generated a sharp DAU spike followed by an equally sharp post-holiday reversal—a pattern that contrasts unfavorably with Qianwen's promotion, which at least connected subsidy spend to real transaction volume. Yuanbao's red-packet mechanics produced engagement without durable behavioral change.

Tencent's response is a studied hedge rather than a pivot. The company has allocated RMB 10 billion into the DeepSeek round, acquiring technology access, talent optionality, and ecosystem integration rights. Simultaneously, on June 8, 2026, Tencent opened WeChat AI interfaces to all Mini Program developers—a distribution decision that routes AI capability into a network of 1.4 billion monthly active WeChat users and over one million Mini Programs.

The strategic implication is that Yuanbao's standalone competitive position may be structurally irrelevant to Tencent's AI outcome. If WeChat AI matures as a service-dispatch layer embedded in existing user workflows, the relevant metric is not Yuanbao's MAU but the volume of AI-mediated transactions flowing through WeChat's infrastructure. Tencent appears to have internalized this logic faster than its external communications suggest.

Kimi Abandons MAU, Achieves ARR Inflection

Moonshot AI's Kimi recorded 7.45 million MAU in May 2026, down 47% year-over-year. Unlike the other declines on the chart, this contraction was deliberate. In a late-2025 all-hands letter, CEO Yang Zhilin formally de-prioritized absolute user volume for 2026, redirecting resources toward Agent subscriptions and enterprise ARR.

The financial logic is straightforward: a non-paying user who generates high daily token consumption is a net negative on a startup's income statement. The cost structure of generative AI products is fundamentally incompatible with the traditional SaaS playbook of acquiring users cheaply and monetizing later.

The pivot produced a measurable inflection. Following the January 2026 launch of Kimi K2.5, cumulative revenue in the first 20 days exceeded full-year 2025 revenue. ARR crossed US$200 million by April 2026. In May 2026, Kimi closed a US$2 billion Series D at a valuation exceeding US$20 billion—approximately four times its valuation from six months prior. Among remaining users, March 2026 average monthly usage frequency reached 23.8 sessions per user, indicating that the retained cohort is more engaged, not less.

Kimi's trajectory is the sector's most direct evidence that monetization pivots are executable—but the durability of enterprise ARR at this growth rate remains unproven over a full business cycle.

Alipay and WeChat Threaten to Render MAU Metrics Obsolete

Ant Group's AI assistant Afu reached 24.12 million MAU, entering the top five. But Ant's more significant move was the June 16, 2026, invitation-only beta launch of an AI-native version of Alipay, which consolidates tens of thousands of life services—provident fund queries, ride-hailing, food ordering, fund purchases—into a single conversational interface.

The threat model this creates for vertical applications is structural: if a user can accomplish the full task loop within Alipay or WeChat without opening a dedicated app, the vertical app's distribution advantage evaporates. Current testing indicates the model performs well on low-complexity decisions (ride-hailing, utility payments) but struggles with high-ambiguity scenarios (open-ended food discovery). The capability gap is real but likely temporary.

The deeper strategic point is that Alipay and WeChat AI are not competing on the MAU dimension at all. They are building service-dispatch infrastructure where the AI interface becomes invisible—embedded in existing workflows rather than standing as a separate destination. If this model matures, monthly active user counts become a lagging indicator of competitive position rather than a leading one.

The Monetization Reckoning Approaches Industry-Wide

The May 2026 QuestMobile data, read in aggregate, maps three distinct competitive strategies converging on the same constraint: compute costs are fixed and rising, free distribution is not a permanent strategy, and user switching costs in AI applications remain structurally low.

ByteDance is routing monetization through Douyin commerce, betting that attention converts to transaction volume. Alibaba is betting that its payment and fulfillment infrastructure makes Qianwen the default purchase-decision interface. Tencent is hedging across DeepSeek infrastructure investment and WeChat AI distribution, effectively conceding the standalone app battle. DeepSeek is building developer ecosystem lock-in through open-source distribution and aggressive API pricing. Kimi has demonstrated that revenue-first pivots are viable but requires sustained enterprise demand to validate.

The AI version of Alipay and WeChat AI represent a fourth model: service infrastructure that monetizes through transaction economics rather than subscription or advertising, and where the user interface itself eventually disappears into the operating system of daily life.

Doubao's May pricing experiment established one data point with industry-wide relevance: even the highest-MAU AI application in the world will absorb meaningful churn when it introduces a paywall into a market conditioned on free access. The compute cost pressure that forced that experiment is not unique to ByteDance. Every company on this chart faces the same underlying equation. The monetization reckoning is not a ByteDance problem. It is a sector-wide event in progress.

Related Coverage:

Chinese Tech Giants Diverge on AI Monetization as ByteDance Breaks the Free Model