The history of Moonshot AI began on March 1, 2023 — the 50th anniversary of Pink Floyd's iconic album The Dark Side of the Moon, which company founder Yang Zhilin considers his favorite. At the time, it was a small IT office reflecting its creator's tastes. Today, three years later, Moonshot AI is preparing for its final funding round before an IPO in Hong Kong, and the company's valuation could reach $50 billion. This is impressive growth that requires detailed analysis.
Yang Zhilin's Path: From Tsinghua to the Top of the AI Industry
Yang Zhilin, born in 1992 in Shantou, is a typical representative of the new generation of Chinese technocrats. He graduated from Tsinghua University, earned a PhD from Carnegie Mellon University, and interned at Google Brain and Meta. Instead of staying in the US, he returned to China and worked on Huawei's PanGu and Wu Dao models. His story is not just the biography of a talented scientist, but a reflection of a systemic shift: China has ceased to be a supplier of talent for American giants and has begun creating its own leaders.
Yang's academic advisor, Ruslan Salakhutdinov, called him "absolutely brilliant" after the launch of Kimi K3. It is important to note that Moonshot AI's co-founders are Yang's classmates from Tsinghua. This is a team built on trust and a shared vision, not on random connections.
Kimi K3: Not Just DeepSeek 2.0
Moonshot AI's financial metrics are staggering. At the end of 2025, the company was valued at $4.3 billion; by February 2026, at $10 billion; in March, at $18 billion; in May, at $20 billion; and by the end of July, at $35 billion. ARR doubled from $200 million to $300 million in just a few months. The next target is $50 billion. Behind these numbers stands a company of 300 people that was virtually unknown outside Asia a year ago.
More than 70% of Moonshot AI's revenue comes from API sales to enterprise clients, with the rest from international Kimi subscriptions. However, after the release of Kimi K3, the company was forced to pause onboarding new users due to a shortage of computing power. This exposed the gap between ambition and physical constraints.
Unlike DeepSeek-R1, which caused a price shock in January 2025 by offering OpenAI o1-level reasoning at 10-30 times lower cost, Kimi K3 does not have such an advantage. Its prices ($3 per 1 million input tokens and $15 per 1 million output tokens) are comparable to Western counterparts like Claude Sonnet 5. This suggests that the price war started by DeepSeek is gradually fading, and competition is shifting to the realm of quality and ecosystem.
Openness as a Strategic Weapon
Initially, Kimi was a closed model, but after DeepSeek's success, the company made a 180-degree turn. On July 11, 2025, Moonshot AI released Kimi K2 — the first Chinese open model with a trillion parameters — and on July 16, 2026, K3 with 2.8 trillion parameters, the largest LLM to date. The model took first place in the Frontend Code Arena ranking, surpassing Claude Fable 5 and GPT-5.6-Sol.
However, K3 has serious issues: the hallucination rate has risen to 51% compared to 39% for K2.6. The model is on average more accurate in its responses but makes errors more frequently. Additionally, running K3 locally requires between 700 GB and 1.6 TB of RAM, making it practically inaccessible to ordinary users.
Moonshot AI's openness is not altruism but a response to competitive pressure. First, it is a way to compete for the global developer market. Second, open weights create a problem for Washington that is difficult to solve legislatively: a model, once released online, is copied to thousands of servers around the world.
The US administration is already considering restrictions on Chinese AI models, while Nvidia, Meta, Microsoft, and 22 other companies have published a letter urging against imposing bans. They warn that restrictions will not strengthen US leadership but will merely shift the market toward a few closed developers.
Sanctions as a Catalyst for Innovation
Washington's export restrictions, intended to slow China's AI development, have had the opposite effect. They forced Chinese labs to seek unconventional solutions. DeepSeek R1 proved that a competitive model can be built without colossal investment, and Moonshot AI continues this line through engineering rigor: aggressive context caching, hybrid attention architectures, and token generation optimization.
State Context
Moonshot AI formally remains a private company, but among its investors is China's National Social Security Fund, which manages pension reserves. The presence of such players is an indicator of the company's significance to Beijing. Preparing for the IPO, the company dismantled its offshore VIE structure, officially recognizing the full jurisdiction of Chinese authorities over its corporate structure and data.
The totality of facts paints a picture of a business that cannot be viewed in isolation from the national context. This is not yet a victory over sanctions, but neither is it their triumph. China is closing the gap through architectural workarounds, not technological superiority. The rules of the game are changing faster than politicians can react to them, and open Chinese models are already being used by developers worldwide.
My analysis: Moonshot AI is not just another Chinese startup, but an indicator of a fundamental shift in the global AI economy. The success of K3 shows that the industry's future will be determined not only by computing power but also by the ability to build open ecosystems. However, the rise in hallucinations and infrastructure constraints serve as a reminder: the AI race is a marathon, not a sprint, and American labs still have time for a countermove.