On June 8th, Bloomberg announced that Moonshot AI is seeking a new round of funding, with plans to raise up to $2 billion and a post investment valuation target of $30 billion.
Equivalent to RMB, approximately 200 billion yuan.
What level is this? Approximately twice the market value of Haidilao, higher than the vast majority of A-share listed companies. And the company that created this valuation is an AI company called “Kimi”, which has only been established for three years, has a total of more than 300 employees, and its product is called “Kimi”.
30 billion US dollars is not even the highest on this track – during the same period, Zhipu had already been listed on the Hong Kong stock market, with a market value exceeding 90 billion US dollars at one point; DeepSeek is reportedly in talks for its first round of external financing, with a valuation of approximately $45 billion; Even MiniMax, which was founded around the same time as Kimi, has landed on the Hong Kong stock market with a valuation of $6.4 billion.
The valuation of Chinese big model companies has collectively taken off in the past six months.
How did these numbers come about? What happened behind the scenes?
01 Termination of an Old Logic
To understand this collective price increase, we need to go back to two years ago.
In 2024, the valuation anchor for large model companies is the number of monthly active users. At that time, Kimi was the one who was most adept at playing with traffic – relying on the differentiated selling point of “2 million words of ultra long text”, the monthly active users once soared to 36 million, which was undoubtedly the top stream of big models at that time. At that time, the industry generally believed that whoever had more users had a future.
This is the valuation inertia left by the Internet era. In the past twenty years, every financing by BAT has told the same story: first bring in users, and then talk about monetization. Investors have been trained by this logic for a whole generation.
But big models are not apps.
The strategy of burning money for traffic has encountered a fatal problem on this track: inference cost. Every user call to the big model is backed by real computing power consumption, and the marginal cost does not approach zero as the user grows, but rather increases. ByteDance, Alibaba, and Tencent can offer almost unlimited subsidies to make their products free, which ordinary startups simply cannot afford.
By March 2026, Kimi’s monthly activity has slipped to 8.34 million, ranking behind Doubao (345 million), Qianwen (166 million), and DeepSeek (127 million), and not even making it into the top five. According to the valuation logic of the Internet, Kimi should have been out long ago.
But its valuation actually rose from $4.3 billion to $30 billion during this period.
What happened in between? The answer is: the rules of the valuation game have been quietly rewritten.
Investors are no longer asking ‘how many users do you have’, but are starting to ask ‘how deep are your technical barriers’ and a more direct question:’ How much money are you making now and what is your growth rate? ‘
Kimi’s answer made the market re-examine it.
In January 2026, Kimi’s paid orders increased by 8280% month on month – more than eighty times. Stripe’s paid ranking jumped from above the top 100 to 22nd globally, and rose to 9th place a month later. By March, the company’s ARR (annualized recurring revenue) exceeded $100 million; In April, it exceeded $200 million – a speed that ranks among the top SaaS companies worldwide.
More noteworthy is where the money comes from. Kimi’s revenue explosion mainly relies on two factors: paid subscriptions and B-end API calls. The latter means that more and more developers and enterprises are using Kimi’s modeling capabilities to build their own products. The well-known code tool cursor has been exposed for calling Kimi’s model at the bottom level – by the way, cursor’s own valuation is also as high as $50 billion.
This is a new valuation logic: the value of big model technology as infrastructure. It is no longer similar to Internet App, but closer to semiconductor or EDA software – the number of users is not necessarily the largest, but every user is inseparable, and is willing to pay real money for it.
02 Dark Moment and Strategic Refocus
But Kimi’s journey to this day has not been smooth sailing. In fact, on the eve of this commercialization outbreak, it was at its most difficult time.
At the beginning of 2025, DeepSeek emerged. Thanks to its low cost and high performance, it quickly went viral both domestically and internationally, becoming a phenomenal event in the AI industry. For Kimi, the impact of this incident came not only from external competition, but also from within – during that time, an employee bluntly stated at the annual meeting: Given the current situation, why do job seekers come to Kimi instead of DeepSeek?
Kimi’s response is a thorough strategic refocusing.
In terms of direction, we will cut off the scattered exploration and fully bet on “model first” – shifting the technological route from “long text assistants” to “agent models for complex tasks”, focusing on three dimensions: token efficiency, long context, and agent swaps. On the product side, Kimi K2.5 was released in early 2026 and immediately became the top call volume on the OpenRouter platform, directly igniting that round of paid growth.
Interestingly, Kimi’s algorithm team’s first reaction to DeepSeek’s popularity was not anxiety, but excitement. A saying circulating internally is: DeepSeek saved us – because it proved the direction of the technological roadmap and forced the team to find its true place.
This mentality may also explain a certain background of Kimi’s company. More than 300 employees, with an average age of less than 30 years old, 80% have an introverted personality, and the company has no departments, job levels, or OKRs. In 2025, a 17-year-old high school intern published a paper as the first author, which was immediately reposted and evaluated by several well-known figures in Silicon Valley. This indirectly proves Kimi’s company culture: as long as the direction is right, age and qualifications are not important.
03 A collective reassessment
Kimi’s story is certainly not an isolated case.
Pulling the timeline to the end of 2025 to the first half of 2026, one will find an interesting thing: the valuations of almost all top model companies are soaring intensively within this window. Zhipu has completed its listing on the Hong Kong stock market and has initiated an A-share IPO; MiniMax also landed on the Hong Kong stock market; DeepSeek is reportedly in talks with institutions such as the National Integrated Circuit Fund for the first round of external financing; It is also rumored that Jieyue Xingchen will soon submit an application for listing on the Hong Kong stock market.
In six months, the valuation coordinate system of this track has been raised by an overall order of magnitude, and the entire industry has reached the same critical point.
This critical point can be described by a set of numbers: by 2026, the monthly active user base of AI native apps in China will reach 440 million, with a quarterly increase of 130 million (QuestMobile data). This means that big model products are shifting from being a “hot topic in technology media” to a daily tool for ordinary users, and the willingness to pay is starting to climb. Kimi’s ARR surged from zero to $200 million within a few months, which is a clear industry signal telling the market that ‘big models can really make money’.
This is a bit like the mobile internet in 2013 and 2014. In those two years, the trigger for market repricing was the emergence of the first batch of truly profitable native apps – WeChat began to commercialize, and Didi ran a unit economy model. Once someone proves that ‘there is really money to be made on this road’, the entire valuation system begins to be rewritten. The trigger for today’s big model race track is the same mechanism, but this time it’s faster and has larger numbers.
Of course, there are also aspects worth examining in this reassessment.
Taking Kimi as an example, ARR of 200 million US dollars is not small, but corresponding to a target valuation of 30 billion US dollars, the PS multiple is about 150 times. This number does not hold water in any mature industry, even in the high growth SaaS arena where top companies typically have PS multiples between 30 and 50 times.
On the other hand, the pressure of burning money is also real. According to data, MiniMax’s monthly cash expenditure is approximately $27.9 million, while Kimi’s size and investment are only slightly larger. The 10 billion yuan cash reserve on the account is not without pressure in the face of continuous model training and computing power investment.
But the other logic that supports these valuations also exists: OpenAI’s valuation reached $300 billion in 2025, and Anthropic’s valuation in June 2026 reportedly surpassed OpenAI, reaching $965 billion. If Chinese model companies can maintain competitiveness in technology and the market size of their services is not inferior to any other economy, whether $30 billion is high or low is a question with no standard answer.
Valuation, ultimately, is an expectation. It measures not today’s revenue, but the collective judgment of the market on where a company can reach at a certain point in the future.
Conclusion
On the day the news of the $30 billion financing came out, neither Kimi nor Yang Zhilin responded.
In the same month, at the end of his speech at the Tsinghua AGI Summit, he shared a conversation with Kimi – he asked Kimi if the arrival of AGI could threaten humanity, and as a researcher, should we continue? Kimi’s answer is: Even if there are risks, it will continue because giving up AGI means giving up the potential of human civilization. Yang Zhilin agreed with that answer and expressed his plan: ‘I hope to continue to do better from K4, K5 to K100 in the next ten or twenty years.’. ”
The game rules have indeed changed. But the game is not over yet, it’s a long journey, Kimi is on the road, and the entire industry is still on the road.
相关文章
