ByteDance’s Doubao Lost 6.1 Million Users the Moment It Hinted at Paid Tiers

The pricing strategy reflects a deliberate segmentation: ¥68 sets a floor for casual upgraders, ¥200 identifies heavy users, and ¥500 is aimed at enthusiasts and enterprise professionals . A Citi survey of 1,800 Chinese users earlier in 2026 found that 45% were willing to pay for advanced AI features — but that the average acceptable monthly price was only ¥48.3, well below Doubao’s entry-level ¥68 .

The immediate fallout: 6.1 million MAUs gone

The backlash was swift. Data from global AI market tracker Aicpb.com showed Doubao’s monthly active users fell by 6.1 million in May — a 1.81% month-on-month decline and a rare slump since the app’s 2023 launch . Analysts at the South China Morning Post described the drop as a “worrying sign,” warning that ByteDance risked “throwing away its lead” by monetizing too early .

When the SCMP surveyed 12 AI users in mainland China, 10 said they were reluctant to pay, citing high prices and perceived performance gaps for work-related tasks . The pattern mirrors earlier failures: Baidu had previously tested paid access for its Ernie Bot only to backtrack and make it free amid intensifying competition .

What the analysts actually said

Nomura was among the first to react, calling Doubao’s subscription launch “a big step forward” with “significant implications for the entire AI chatbot market in China” . But Nomura’s optimism came with a hard historical caveat — the firm highlighted that the Chinese consumer market has long been “a paid-subscription graveyard” for general-purpose apps .

Morgan Stanley viewed the initiative as a positive signal for the industry’s path toward commercial sustainability. The key insight: with approximately 345 million monthly active users at the time, Doubao didn’t need a high conversion rate to generate meaningful revenue . The firm’s analysts noted that “maintaining flat free access for heavy workloads” was economically unsustainable, and the tiered plans “effectively asked heavy users to subsidise their own compute” .

Industry observers quoted by China Daily and other outlets saw the subscription test as an industry-wide pivot — the end of a “burning cash for daily active users” model that had defined China’s large-model landscape . With inference costs rising in lockstep with usage, the free-for-all model was becoming untenable .

Skeptical voices from outlets including the SCMP and KR Asia pointed to the immediate user exodus as evidence that Chinese AI users are not yet ready to pay. One 36Kr source noted that Doubao would not use paid-user penetration as a performance metric in 2026 — a tacit acknowledgment that ByteDance itself wasn’t expecting quick adoption .

A fragile lead and an expensive bet

The subscription rollout was a necessary risk: ByteDance needed to test whether its massive free user base could be converted into a revenue engine before compute costs overwhelmed the model. The early returns didn’t suggest an easy path. But Morgan Stanley’s framing — that high user volume compensates for low conversion — might be the bet ByteDance is actually making.

What the test proved is that even China’s largest consumer AI platform can bleed users the moment it asks for money. Whether Doubao’s premium productivity features can change the country’s aversion to paying for software — a cultural pattern that has sunk similar efforts before — remains an open question .