
From questioned spending on user acquisition, to Kimi K3 shaking up Silicon Valley, to a confidential IPO filing — Yang Zhilin has moved Moonshot AI a long way in less than two years.
On September 2, according to LatePost, Moonshot AI confidentially submitted an A1 listing application to the Hong Kong Stock Exchange earlier this week, formally initiating the process for a potential Hong Kong IPO. CICC and Goldman Sachs have reportedly been appointed as joint sponsors. Moonshot AI said it currently has no information to disclose.
If the plan moves forward, the company could become the third major Chinese foundation-model company to go public, following Zhipu AI and MiniMax, potentially in early 2027. Moonshot AI has taken just over three years to move from founding to an IPO process.
That pace is particularly striking given that as recently as late 2025, founder Yang Zhilin said in an internal letter that the company was not in a hurry to go public.
Eight months later, the strategy appears to have changed dramatically. At the center of that shift is Kimi K3.
Moonshot AI's private-market valuation has surged dramatically in 2026.
The company was valued at approximately $4.3 billion after completing its Series C financing in December 2025. The pace accelerated sharply in 2026, with three consecutive funding rounds between January and February pushing its valuation from $10 billion to $18 billion.
Following its Series D in May, the valuation reached $20 billion. By June, its Series E round reportedly lifted its pre-money valuation to $31.5 billion. In July, the company raised more than $3.5 billion in Series F financing, bringing its post-money valuation to $35 billion.
Its planned August G round, positioned as a pre-IPO financing, was subsequently brought forward, with the company's pre-money valuation reportedly reaching $50 billion.
In just eight months, Moonshot AI's valuation increased more than tenfold.
The most direct catalyst for this valuation surge was the launch of Kimi K3 in July.
With 2.8 trillion total parameters, native visual understanding, and a 1-million-token context window, K3 was designed for complex tasks including software engineering, knowledge work, deep research, and multimodal reasoning.
In the Frontend Code Arena benchmark, K3 reportedly reached the top of the global leaderboard with a score of 1,679. In Artificial Analysis' overall evaluation, it ranked third globally, behind Claude Fable 5 and GPT-5.6 Sol.
The market reaction was immediate. On the day K3 was released, Zhipu AI's stock fell 28.49%, while MiniMax dropped 15.62%. The episode was widely compared by overseas media to a potential "DeepSeek 2.0" moment.
More importantly, Moonshot AI's commercial performance has been accelerating.
Its annual recurring revenue (ARR) surpassed $100 million in March 2026, reached $200 million in May, and exceeded $300 million in June. API revenue reportedly accounts for more than 70% of total revenue, with overseas developers becoming an increasingly important customer base.
Behind the steep valuation curve, however, lies a period of intense pressure.
When DeepSeek R1 emerged in January 2025, it fundamentally challenged the prevailing growth playbook for Chinese AI applications. DeepSeek relied heavily on model performance and organic adoption rather than large-scale advertising, forcing companies such as Moonshot AI to rethink their approach.
Moonshot AI subsequently halted consumer-focused advertising and suspended several other C-end applications, including Ohai and Noisee. Its user-acquisition spending reportedly fell by more than 70% from the fourth quarter of 2024, while the company entered a financing gap that lasted roughly 16 months.
Yang Zhilin then made two critical strategic decisions.
First, Moonshot AI moved away from the increasingly expensive competition for consumer traffic and went all in on foundation-model development.
Second, after DeepSeek established a strong position in reasoning, Moonshot AI shifted its focus toward AI agents, moving closer to the technology direction pioneered by Anthropic.
In February 2025, Moonshot AI released Kimi K2, described as the world's first open-source trillion-parameter MoE model, with strong performance in coding, agentic tasks, and mathematical reasoning.
K2.5 subsequently introduced agent-cluster capabilities, coinciding with the rapid rise of agentic AI. Less than 20 days after its release, revenue reportedly exceeded the company's total revenue for 2025.
Then came K3 in July 2026, with the company's roadmap around hybrid linear attention, attention residuals, and RL scaling largely materializing in the new model.
Three years after its founding, Moonshot AI's shareholder base reportedly includes leading investors such as Alibaba, Tencent, Meituan Longzhu, Sequoia China, and IDG, alongside strategic investors including China's social security fund and China Mobile's industry fund.
Technology breakthroughs, surging fundraising, and a dramatic valuation increase have all converged within a remarkably short period.
But reaching the summit is not the same as securing the win.
Yang Zhilin once compared building large AI models to climbing a snow-covered mountain: perhaps one day, he said, we will discover that the mountain has no end — and he hopes it never does.
AI Market Research & Consumer Insights Provider
China's Leading Treasury Management Software Provider
AI落地解决方案及数字化创新服务提供商
A high-tech enterprise integrating product design, R&D, sales, and services.
Leading full-scenario data intelligence service provider in China.
