As 2025 draws to a close, the AI sector is presenting a stark picture of two extremes.
On one side, companies like Zhipu AI and MiniMax have successively passed hearings at the Hong Kong Stock Exchange, making a full sprint to become the "first AI large model stock"; on the other side, Moonshot AI, also one of the "AI Six Tigers," is deeply embroiled in a public opinion vortex due to declining weekly active users and its industry ranking.
According to the latest report from Quest Mobile, the weekly active user count for Moonshot AI's Kimi product has recently dropped to 4.5 million, and its ranking has fallen from second place a year ago to seventh, being surpassed by Doubao, DeepSeek, Yuanbao, Ant's A-Fu, and Alibaba's Tongyi Qianwen.
Furthermore, data from Qimai shows that since April of this year, Kimi's overall download volume has experienced a significant decline and has remained at a low level for an extended period. Behind this, the weakness of its technological moat and the failure of its "burning money for growth" model are likely the primary reasons for the drop in downloads and daily active users.
An AI industry investor, in communication with the BUG column, pointed out that the biggest problem facing Moonshot AI currently is: "What it has (in terms of product or technology), others have too, but it's not the best."
In early 2024, Moonshot AI rapidly sparked widespread discussion online thanks to the exceptional long-context processing capability of its Kimi Smart Assistant. It subsequently secured a massive investment exceeding $1 billion from Alibaba, briefly becoming the most promising "AI star" startup in China.
After receiving substantial funding from Alibaba and other institutions, Moonshot AI's marketing efforts became "frenzied." Data indicates that its monthly advertising spending peaked at nearly 200 million yuan, once creating a "screen-dominating" trend on platforms like Bilibili. This "money-splashing" marketing tactic pushed Kimi's monthly active user base to over 36 million around October last year.
Amidst this rapid growth, internal governance issues arising from a conflict of interest between co-founder Zhang Yutong and Zhu Xiaohu, Managing Partner at GSR Ventures, coupled with intense external competitive pressure from major players like Doubao swiftly fighting back, caused the technological halo around Moonshot AI's Kimi to fade, and its market position began to slide.
According to the "Report on AI Application Interaction Innovation and Ecosystem Implementation in the Second Half of 2025" recently released by QuestMobile, in the latest statistical period (December 8-14, 2025), Doubao led the domestic AI-native App weekly active user ranking with 155.2 million users, followed by DeepSeek with 81.56 million, and Yuanbao with 20.84 million. Ant's A-Fu and Alibaba's Tongyi Qianwen ranked fourth and fifth respectively. Kimi, which ranked second last year, fell to seventh place with only 4.5 million weekly active users, relegating it to a second-tier large model product.
Prior to this, data from QuestMobile's "Q3 2025 AI Application Value List" showed that the monthly active user scale of Kimi Smart Assistant also declined from 14.072 million in Q2 2025 to 9.926 million in Q3 2025, a sequential decrease of approximately 30%.
The BUG column's query via Qimai data revealed that regarding download volume, after experiencing rapid growth in February this year, Kimi's overall downloads have seen a significant decline since April and have remained at a low level for an extended period.
Behind the dual decline in downloads and user activity, the loss of technological leadership and the failure of the "burn money for growth" model are the most significant reasons.
In 2024, Kimi built a phase-leading advantage with its killer feature of "long-context processing," accurately addressing the pain points of professionals in areas like million-word contract analysis and long research report interpretation, quickly making it a capital-favored "AI rising star." However, this technology was soon matched and surpassed by leading giants like ByteDance and Alibaba, directly undermining Kimi's vaunted "long-context processing" advantage.
AI industry investor Li Yang (pseudonym) stated bluntly in communication with the BUG column: "Long-context processing is not rare in tech circles; anyone can do it if they want. The main reason only Kimi pursued it initially was because long-context consumes immense computing power, the cost is too high, and without suitable application scenarios, no one else was willing to do it."
Once the technological barrier built on "long-context processing" was breached, the only drivers for Kimi's user growth were "burning money for user acquisition" or creating new "technological differentiation barriers." However, judging from the past year, although Kimi briefly established leading technological advantages by releasing models like K2, these were quickly nullified or even surpassed by players like OpenAI, Google, as well as DeepSeek, Alibaba, and Zhipu AI, making it impossible to maintain a long-term technological lead.
On the path of "burning money for user acquisition," DeepSeek's explosive growth story of "reaching 100 million users in 7 days," achieved through technological breakthroughs, directly declared the "burn money for growth" model for AI applications as inefficient and unsustainable.
Li Yang revealed that during periods of aggressive advertising spending, Kimi's cost per acquired user was around 10 yuan. Factoring in the computing costs incurred from user queries or long-context processing experiences after acquisition, the comprehensive customer acquisition cost per user was approximately 12-13 yuan. If 200,000 new users were added daily, Kimi would burn through about 2.5 million yuan per day. "If these users cannot be converted into paying customers, long-term losses are an inevitable result."
Given the current situation of only 4.5 million weekly active users and an industry ranking dropped to seventh, a significant portion of the users acquired through Kimi's early high-cost spending have already churned or become "dormant." In contrast, DeepSeek, which rarely engages in paid user acquisition, still boasts 81.56 million weekly active users and has taken over the industry's "second place" position previously held by Kimi.
Li Yang stated directly, "The most awkward situation for Moonshot AI currently is that what it has, others have too, and it's not even the best."
As its differentiated technological advantage vanishes, and the burn-money-for-growth model proves ineffective, Moonshot AI's current predicament is becoming increasingly awkward, caught between the fierce encirclement by major players like Doubao, Yuanbao, and Tongyi Qianwen, and the chasing pressure from other "AI Large Model Six Tigers" like Zhipu AI and MiniMax.
The BUG column notes that regarding its business model, Moonshot AI currently monetizes its C-end primarily through tipping and subscription fees for the Kimi Smart Assistant, while the B-end is monetized mainly through large model API calls. However, for C-end monetization, nearly all features currently offered as paid items in Kimi are available for free through products like Doubao, Quark, Tongyi Qianwen, and Lingguang. This means that in a market where users have weak payment awareness and numerous alternative products exist, as long as the major players do not charge fees, it is difficult for Kimi to retain paying users long-term, making churn inevitable. For the B-end monetization model, aside from API calls, Moonshot AI's progress in areas like customized development and major client cooperation partnerships is somewhat weaker compared to giants like ByteDance and Alibaba, and it also lags slightly behind startups like Zhipu AI.
In the view of Li Mingshun, Chairman of Hanghang AI, globally, as scaling laws reach their peak, the capability bottlenecks of AI large models are becoming increasingly apparent, and the entire industry has entered a "cards-on-the-table" phase. In this stage, large companies have stronger advantages than small ones, and entrepreneurs should focus more on finding specific, strong application scenarios to create fully closed-loop AI application products, rather than building broad, all-encompassing products.
However, judging from the product forms pursued by Moonshot AI, both its C-end and B-end products exhibit high overlap with those of leading internet giants. It has not, like Baichuan AI, focused early on healthcare, nor has it, like Zhipu AI, intentionally targeted B-end and G-end (government) sectors, and it certainly hasn't, like 01.AI, abandoned the trillion-parameter large model race to embrace partnerships with firms like DeepSeek.
A veteran AI practitioner直言不讳地 told the BUG column that Moonshot AI's current situation is a clear case of being "caught between a rock and a hard place."
Advancing, by working on large models, makes meaningful innovation and improvement difficult; retreating, by shifting to applications, would "sacrifice" its high valuation. This predicament also reflects the common situation faced by the broader AI startup community: on one hand, they aim high for greater opportunities, but lacking sufficient market strategic insight inevitably leads to setbacks.
In his view, perhaps Moonshot AI should promptly steer away from the main course already targeted by giants, and instead choose more vertical, focused scenarios to develop more distinctive features, or pursue globalization earlier to enter the broader global market.
Comments