Kimi Goes Viral: Who Sparked Another Frenzy in Large Models?
-
From last year's excitement to this year's calm, despite never lacking attention and traffic, China's large model and AI circles haven't seen such a frenzy in a long time.
Over the past week, the ripple effects caused by AI large model startup Moonshot AI and its product Kimi have spread from social media to the A-share market, driving a surge in so-called 'Kimi concept stocks.' Zhuang Minghao, a well-known internet investment analyst, even exclaimed on Jike: "Seriously? Kimi can drive stock trends in the A-share market? This is just too abstract."
In fact, since the emergence of ChatGPT, similar hype cycles have been common. Whether it was the laughable 'daughter concept stocks' triggered by the viral success of text-to-video startup Pika late last year or this year's 'Sora concept stocks,' the AIGC trend remains strong. At least based on the current reactions in the capital markets, optimism still dominates. Setting aside the narrative of "Kimi driving the A-share market," the explosive popularity of MoonDark and its product Kimi reflects at least two definitive trends: First, MoonDark has always been a star company in the AI circle, so Kimi's sudden fame isn't entirely short-term hype but rather the evolution of competition among leading AI model startups, driven by capital, products, and other multidimensional factors. Second, competition in the application layer of large models will continue to intensify. Unlike previous short-lived AI-native applications like "Miao Ya," Kimi, as a relatively pure productivity tool, carries significance and value that may align more closely with people's expectations for a "Killer APP" in the AI era.
However, the road ahead for Kimi is far from smooth. Its key product and technological differentiator—"long-text" capabilities—isn't entirely novel. After the initial hype, how will Kimi sustain its momentum? How will it approach commercialization? These are the critical areas Kimi needs to focus on and invest in beyond just managing its sudden popularity. Before Kimi captured nationwide attention, its parent company Moonshot AI was already a star enterprise in the large language model (LLM) circle. Founded by AI prodigy Yang Zhilin, Moonshot AI was recognized by tech media The Information as one of the potential "Chinese OpenAI" contenders even before releasing any products. Notably, Yang Zhilin was the only individual listed by name in this prestigious company ranking.
The combination of technical talent and the aggregation effect has attracted significant investment to Moonshot AI under the Chinese venture capital logic that "investing in large models means investing in people." In February this year, Moonshot AI completed a Series B funding round exceeding $1 billion, with investors including Alibaba, Meituan, Xiaohongshu, and Sequoia Capital China. According to LatePost, Moonshot AI's post-money valuation has reached approximately $2.3 billion, making this the largest single funding round secured by any domestic AI large model company to date. However, unlike the previous self-admiration in the AI circle, Kimi's current popularity on social media and even in the A-share market is more driven by traffic logic—"Everyone is talking about Kimi." A netizen commented on social media. The surge in traffic has also led to website anomalies such as downtime for Kimi. On March 21, Moon Dark Side issued a statement saying: "We have observed that Kimi's system traffic continues to rise abnormally, and the trend of traffic increase far exceeds our resource planning expectations."
"Silicon Research" reviewed Kimi's popularity and found that it was actually the result of timing, location, and human factors working together. First, the timing is right. Kimi's performance in the consumer market is on par with that of well-funded, high-traffic industry leaders, showcasing its potential as a "Killer APP."
The development of large models has reached two critical junctures: First, the competition around model layers, once centered on "parameter supremacy," has gradually evolved into a battle for user acquisition at the application layer. Robin Li, founder of Baidu, has repeatedly emphasized that "competing over models is less valuable than competing over applications." Second, products from leading large model companies have now been made available to the public, entering a phase of product diffusion. This means the effectiveness of these products will be directly reflected in user feedback and specific product metrics. Under two major nodes, over the past period, whether it's Alibaba's Tongyi Qianwen or Baidu's Wenxin Yiyan, major tech companies have been actively promoting their AI tools through their own channels and leveraging traffic resources. However, despite their 'brute force for miracles' approach, their results have been far from remarkable compared to ChatGPT's achievement of reaching 100 million monthly active users in just two months. According to QuestMobile data from January 2024, only Douyin's Doubao and Baidu's Wenxin Yiyan surpassed 10 million monthly active users (Doubao with 17.57 million and Wenxin Yiyan with 12.13 million).
In contrast to these major players, the relatively unknown Kimi has demonstrated a 'surpassing the giants' rapid growth trend. Citing data from the February AI product rankings released by aicpd.com, Kimi, ranked third, showed a listing visit growth rate of 107.6%, while the top-ranked Wenxin Yiyan and second-ranked Tongyi Qianwen both experienced varying degrees of decline in their listing visit growth rates. Apart from 'right timing', Kimi and Moon Dark Side's first-mover advantages have collectively erupted in this round, which represents 'geographical advantage'.
Different from similar AI dialogue products, what exactly are Kimi's characteristics and differentiators? 'Silicon-Based Research Lab' has identified three main aspects: positioning, user operations, and marketing promotion: First, in terms of product positioning, Kimi has been focusing on "long text + To C" from the beginning. This combination directly addresses users' needs for "AI search + document summarization," while the "long text format + source display" approach enhances the reliability of responses to some extent.
Second, it implements multi-platform user operations. We observed that Kimi is one of the early large model companies to deploy across web, app, and WeChat mini-programs. This dual approach of lightweight and comprehensive platforms enables Kimi to iterate faster based on user feedback. 三是推广裂变。从长文本的使用场景来看，早期产品使用者集中在科研、法律、金融、媒体等垂直领域，而这部分垂直领域的KOL为Kimi带来了自然流量与口碑积累。另一方面，据「硅基研究室」观察，Kimi也在近期加大了营销投放力度，发力获客。
更为关键的，还有人和——如「东四十条资本」的评价：“月之暗面对各种节奏的把控相当老辣，甚至很难相信这是家仅成立一年的创业公司。” After announcing a $1 billion funding round in mid-February, several securities firms began publishing research reports on Kimi. In March, Tencent Technology's two exclusive interviews with Yang Zhilin and Zhu Xiaohu, managing partner at GSR Ventures, categorized China's major AI model players into 'technology believers' and 'market believers.' Yang Zhilin, representing the former, stated in the interview: 'AI is not about finding PMF (Product-Market Fit) in the next one or two years, but about how it will change the world in the next ten to twenty years.' Shortly after, on March 18, Moonshot AI announced that Kimi's context input length had expanded from 200,000 words in October last year to 2 million words.
From this perspective, Kimi's explosive popularity is no accident—it is both the inevitable result of its technological accumulation and commercialization efforts and a manufactured inevitability.
In 2024, viewed by many AI optimists as the 'Year of AI Applications,' Kimi's breakout success undoubtedly validates this hypothesis. Huafu Securities pointed out that Kimi may propel domestic AI applications into the 'hundred-model inference' era. Galaxy Securities also mentioned in its analysis report: Kimi Smart Assistant is another 'milestone' in the AGI process, heralding the official entry of large models into the 'long-text era.'
The tech industry excels at storytelling, but no one seems willing to miss the Kimi wave.
According to incomplete statistics from 'Silicon Research Lab,' over the past week, more than ten listed companies, including BXRJ, SDIC Intelligence, Yidian Tianxia, Insig Group, Sangfor, Jiuan Medical, Chaoxun Communication, Hand Information, Zhangyue Technology, Wanxing Technology, and Shengtian Network, have responded on platforms like Interactive Easy that they have collaborations or equity stakes with Kimi. The aforementioned companies' direct or indirect relationships with Kimi also reflect the external expectations for this large model enterprise's capabilities and commercialization. Kimi's sudden popularity, who exactly benefits from it? Or, in comparison to OpenAI's rise, as domestic AI large model unicorns begin to step into the spotlight, who can reap the benefits brought by Kimi's ascent?
From a business and investment perspective, behind Kimi and Moon's Dark Side stands Alibaba, which is more aggressive and dominant in the era of large models. Among China's current AI unicorns, Alibaba has invested in Baichuan Intelligence, 01.AI, Zhipu AI, and Moonshot AI, with its investment in Moonshot AI being Alibaba's largest in the large language model (LLM) sector to date. According to a report by "Dark Surge," a dollar-fund AI investor explained their decision to withdraw from Moonshot AI: "Alibaba increased Moonshot's valuation by 50% and took a 40% stake, so we ultimately backed out."
The reason behind Alibaba's heavy investment in Moonshot AI lies in computing power. Mirroring the "OpenAI + Microsoft Azure" collaboration model, LLM companies require computing power while cloud providers can supply it. Under this partnership framework, LLM enterprises alleviate their financial burdens related to computing power, while cloud vendors can better market their cloud services through AI, seizing opportunities in AI computing infrastructure. Beyond Microsoft Azure and OpenAI, similar models exist in the collaborations between Anthropic and Google, as well as Amazon.
After its explosive popularity and breakout success, Kimi certainly doesn't face a smooth path ahead. The most concerning issue is Kimi's commercialization strategy. Currently, large model companies mainly have three revenue models: First, targeting C-end users with membership value-added services, charging users monthly/yearly on the application side. Compared to Wenxin Yiyan's trial of paid services, Kimi currently remains in a free model. Second, selling APIs, charging companies or developers based on usage frequency, which is a common practice among many model companies including Kimi. Third, selling B-end solutions, such as launching industry/vertical-specific large models.
Unlike companies like Baichuan Intelligence and Zhipu AI that adopt both To B and To C approaches, Kimi has been firmly focused on To C from the beginning and aims to become the 'Super App' of the AI era. Although its direction is clearer, detailed commercialization plans have not been disclosed. Another challenge related to commercialization is user strategy. Behind the surge in visits, Kimi faces the dual challenges of acquiring and retaining customers. According to Sina Tech, the current cost of acquiring customers through advertising for Kimi is approximately 10 RMB. When factoring in the computing costs generated by user interactions after acquisition, the cost per user rises to 12-13 RMB. Whether these costs can effectively translate into user retention will test Kimi's subsequent product iterations and operational capabilities.
Not long ago, major companies such as Alibaba, 360, and Baidu announced the opening of long-text capabilities. For Kimi and Moon Dark Side, a new round of challenges has already begun. In this round of competition, will Kimi maintain its lead and become a true 'Super APP,' or will it be a flash in the pan? Undoubtedly, there are still many questions left for this star company to answer.