Cash Burned, Talent Gone… Has Once-High-Flying Kimi Slipped into the Second Tier a Year Later?

As 2025 draws to a close, the AI race is splitting into two very different realities.

On one side, Zhipu and MiniMax have moved through the Hong Kong listing hearing process, sprinting toward the coveted label of “the first publicly listed large-model company.” On the other, Moonshot AI—also part of the so-called “AI Six Tigers”—has found itself pulled into a storm of criticism as its weekly active users and industry ranking slide.

According to QuestMobile’s latest report, Kimi’s weekly active users have recently fallen to 4.5 million. Its position has dropped from No. 2 a year ago to No. 7 today—overtaken by Doubao, DeepSeek, Yuanbao, Ant’s “Afu,” Alibaba’s Qwen, and others.

Data from Qimai also points to a sharp decline in downloads since April, with Kimi’s overall download volume staying at a relatively low level for an extended period. Behind the numbers, two forces appear to be converging: weaker-than-expected technical moats, and a “spend money to buy growth” playbook that no longer works as it once did.

As one AI investor told the BUG column, Moonshot AI’s biggest issue right now is simple and brutal: “What it has, others also have—but it isn’t the best at any of it.”

Downloads and activity both slide

In early 2024, Moonshot AI surged into the spotlight thanks to Kimi’s standout long-context capability, sparking widespread discussion across Chinese social media. Not long after, it reportedly secured more than US$1 billion in investment from Alibaba, and quickly became one of the most celebrated “AI star” startups in the country.

After raising large rounds, Moonshot AI’s marketing efforts became noticeably more aggressive. At its peak, monthly ad spend reportedly approached RMB 200 million, creating a near “everywhere you look” presence across platforms like Bilibili. That high-intensity spending helped push Kimi’s monthly active users past 36 million around October of last year.

But rapid growth also brought problems. Internal governance tensions—sparked by controversy around a perceived conflict of interest involving co-founder Zhang Yutong and GSR Ventures’ Zhu Xiaohu—combined with fierce pressure from big-tech challengers such as Doubao. Kimi’s early technical halo began to fade, and its market position started to erode.

QuestMobile’s 2025 H2 AI App Interaction Innovation & Ecosystem Deployment Report shows that in the latest measured period (Dec 8–Dec 14, 2025), weekly active user rankings among China’s native AI apps were led by Doubao (155.20 million), followed by DeepSeek (81.56 million) and Yuanbao (20.84 million). Ant’s Afu and Alibaba’s Qwen took fourth and fifth. Kimi, which ranked second last year, fell to seventh—with weekly active users at just 4.5 million—effectively slipping into “second-tier” status.

Earlier, QuestMobile’s 2025 Q3 AI Application Value Ranking also indicated that Kimi’s monthly active users declined from 14.072 million in Q2 2025 to 9.926 million in Q3, a roughly 30% quarter-over-quarter drop.

On the download side, BUG’s review of Qimai data suggests that after a burst of momentum in February, Kimi’s downloads dropped sharply starting in April and have remained subdued since.

Why the “cash-burn” playbook stopped working

The simultaneous decline in downloads and engagement points back to two core issues: the loss of technical edge, and the diminishing returns of paid user acquisition.

In 2024, Kimi built a temporary lead through “long-context processing,” a compelling advantage for scenarios like analyzing million-word contracts or interpreting long research reports—use cases that strongly resonated with professionals. But that advantage was quickly matched and, in some cases, surpassed by tech giants such as ByteDance and Alibaba, weakening the differentiator Kimi was most known for.

An AI investor using the alias Li Yang told BUG that long-context processing was never inherently rare: “If you want to do it, you can.” The reason few pushed hard at first, he explained, was cost—long context consumed enormous compute, and without obvious high-value scenarios, many players didn’t see the payoff.

Once Kimi’s long-context moat was breached, the levers for growth narrowed to two options: continue buying traffic at scale, or build a new, defensible technical differentiation. Over the past year, Kimi did briefly regain momentum with releases such as K2, but the gap didn’t last. Competitors—including OpenAI, Google, and domestic players like DeepSeek, Alibaba, and Zhipu—rapidly closed in or pulled ahead, making sustained technical leadership difficult to maintain.

Meanwhile, DeepSeek’s breakout narrative—massive user growth driven primarily by perceived technical leaps—has further highlighted how inefficient and fragile pure “spend money to grow” can be in AI apps.

Li Yang also shared a concrete cost picture. At the most aggressive traffic-buying stage, Kimi’s cost to acquire a new user was around RMB 10. Add the compute costs from onboarding usage—Q&A, long-document processing, and other high-token activities—and the all-in acquisition cost could reach RMB 12–13 per user. If the app added 200,000 new users per day, that would mean roughly RMB 2.5 million burned daily. “If those users don’t convert to paying customers,” he said, “long-term losses become inevitable.”

Judging by today’s 4.5 million weekly active users and a No. 7 ranking, a large portion of the users originally purchased through expensive traffic campaigns have either churned or gone dormant. In contrast, DeepSeek—widely seen as relying far less on paid acquisition—now holds 81.56 million weekly active users and has effectively replaced Kimi’s former No. 2 position.

Li Yang summed it up bluntly: “Moonshot AI’s awkward reality is that what it has, others also have—and it isn’t the best.”

Stuck between advancing and retreating

When differentiation fades and paid growth stops paying off, the squeeze becomes intense. Kimi now faces pressure from big-tech AI products such as Doubao, Yuanbao, and Qwen, while also being chased by peers like Zhipu and MiniMax. Moonshot AI’s position has become increasingly uncomfortable.

In monetization, Moonshot AI’s consumer business largely relies on tipping and subscriptions via Kimi, while its enterprise side charges mainly through large-model API usage. The problem on the consumer side is that most paid features in Kimi can be obtained for free via competing products such as Doubao, Quark, Qwen, and others. In a market where willingness to pay is limited and substitutes are abundant, Kimi’s paid users are hard to retain as long as major platforms keep similar capabilities free.

On the enterprise side, Moonshot AI appears to have made less visible progress—beyond API access—in areas like custom development and key-account expansion, compared with giants like ByteDance and Alibaba, and it may also lag some startup peers like Zhipu in certain commercial pushes.

Li Ming Shun, chairman of Hanghang AI, argues that as scaling laws show signs of peaking, capability bottlenecks are becoming more obvious—and the industry is entering a “cards-on-the-table” stage. In that phase, large platforms naturally gain an edge. For startups, he suggests, the smarter move is to find strong, narrow scenarios and build end-to-end AI application loops, rather than competing head-on with broad, general products.

From Moonshot AI’s current product shape—both consumer and enterprise—its overlap with top internet platforms remains high. It doesn’t resemble Baichuan Intelligence’s early focus on healthcare, nor Zhipu’s clearer tilt toward B-end and government business, nor 01.AI’s earlier decision to step away from massive-parameter ambitions and embrace ecosystems like DeepSeek’s.

One veteran AI practitioner told BUG that Moonshot AI is facing a classic “advance or retreat” dilemma.

Advance—and keep betting on foundational models—and innovation becomes harder and costlier. Retreat—and pivot fully into applications—and it risks sacrificing the high valuation narrative that comes with being a “frontier model” company. In many ways, this mirrors the broader startup reality in AI: ambition is necessary, but without sharper strategy, the market is unforgiving.

In his view, Moonshot AI may need to step away from the main lanes already locked up by giants, move into more vertical and focused scenarios, develop more distinctive capabilities, or pursue globalization earlier—opening up a far broader market than the current domestic battlefield.