Từ 300 triệu lên 35 tỷ, Moonshot AI gõ cửa bàn chơi AI đẳng cấp thế giới với mô hình mã nguồn mở

Original Author: Sleepy

Kimi has just opened the model weights for K3, and Moonshot AI's new financing round has also been finalized.

According to an exclusive report from a reporter at the Science and Technology Innovation Board Daily, Moonshot AI has completed its Series F financing, raising over $3.5 billion, with a post-money valuation reaching $35 billion.

This financing round was not originally planned to raise this much money. The report states that due to investor subscription amounts exceeding the original target by more than three times, Moonshot AI closed the Series F round early. Consequently, the Series G round, originally scheduled to launch in August this year, has also been brought forward.

Series G will be Moonshot AI's Pre-IPO round before its public listing, with the pre-money valuation already risen to $50 billion.

Just a week ago, the market rumor was that Moonshot AI would complete a round of financing with a pre-money valuation of about $31.5 billion, followed by a final private financing round in August. Now, the Series F round has ultimately raised over $3.5 billion, resulting in a $35 billion post-money valuation, and the next round is no longer waiting until August. Both the money and the valuation have arrived faster than originally planned.

Since its founding in 2023, Moonshot AI has taken just over three years to push its valuation from $300 million to $35 billion. The ongoing Pre-IPO round has already affixed the next price tag at $50 billion.

Over a dozen financing rounds in three years, valuation surges from $300 million to $35 billion

Moonshot AI was founded in April 2023 by Yang Zhilin, Zhang Yutao, Zhou Xinyu, and Wu Yuxin.

About two months after its establishment, the company completed an angel round of over $200 million, with a post-money valuation of approximately $300 million. For a startup large model company whose product had not yet been officially launched, this was already a remarkably large early-stage round.

What truly propelled Moonshot AI to the center of the capital table was its Series A+ round completed in February 2024.

This round raised over $1 billion, led by Alibaba, with participation from institutional and industrial capital such as Sequoia Capital China, Xiaohongshu, and Meituan, pushing Moonshot AI's valuation to around $2.5 billion. Six months later, the Series B round raised over $300 million, increasing the company's valuation further to $3.3 billion.

However, after that 2024 financing round, Moonshot AI's fundraising pace slowed down for a time.

During that year, the Chinese large model market underwent significant changes. Major players like ByteDance and Alibaba continued to drive down model prices. DeepSeek rapidly rose with its open-source approach and cost efficiency. Competition among general-purpose chat products gradually shifted from user growth to model capability, inference cost, and commercial revenue. Kimi initially stood out with its long-context capabilities, but relying on a single product feature was no longer sufficient to sustain a higher valuation.

It wasn't until the end of 2025 that Moonshot AI completed a $500 million Series C round, achieving a post-money valuation of $4.3 billion. After that, the company's fundraising pace noticeably accelerated.

In the first two months of 2026, Moonshot AI completed multiple financing rounds consecutively. Its valuation climbed from the previous $4.3 billion to $10 billion, and further to $18 billion.

In May of this year, the company completed a Series D round of approximately $2 billion, reaching a post-money valuation of $20 billion. Participants were no longer limited to internet companies and market-oriented investment institutions; China Mobile, Guozhi Investment, CPE Yuanfeng, and several state-backed funds also joined the shareholder list.

Shortly after the Series D round was completed, a new round was initiated in June. At that time, the market-rumored pre-money valuation had already reached $31.5 billion. Now, this round has concluded with a final financing amount of over $3.5 billion, raising Moonshot AI's post-money valuation to $35 billion.

Looking back at this financing trajectory, the most dramatic changes have occurred in the past six months.

At the end of 2025, Moonshot AI's valuation was still $4.3 billion. Just over half a year later, this figure has reached $35 billion, an increase of more than seven times. If the next round is completed at a $50 billion pre-money valuation, Moonshot AI's valuation growth over three years will exceed 160 times.

The rationale from investors is also becoming more direct. Previously, they were betting on Yang Zhilin and a Tsinghua-affiliated technical team; now, they are betting on whether Moonshot AI can become one of the few Chinese companies remaining on the global frontier model table.

After K3, Moonshot AI returns to the center of the table

The timing of this early financing closure coincides precisely with the release of Kimi K3.

On July 16th, Moonshot AI released Kimi K3. On July 27th, the company further opened the complete model weights, technical report, and some key infrastructure technologies for K3.

K3 is a Mixture of Experts (MoE) model with a total of 2.8 trillion parameters, activating 104 billion parameters per inference. It supports a context window of 1 million tokens and possesses capabilities in vision, coding, reasoning, and long-duration tool invocation.

In its technical report, Moonshot AI stated that K3 achieves frontier performance on tasks such as long-context coding, agent tasks, knowledge, reasoning, and vision. Its overall capability is very close to the strongest closed-source models, Claude Fable 5 and GPT-5.6 Sol.

Leaderboards change frequently, and one should also maintain a degree of skepticism towards model companies' own tests. The more significant impact of K3 is that it pushes the boundary of open-weight model capabilities further forward.

For a long time, open-source models mostly played the role of chasers. Closed-source companies were responsible for training the strongest models, and six months or a year later, the open-source community would replicate some of their capabilities at a lower cost.

K3 changes this time lag. A Chinese company has directly opened up a model approaching the world's most advanced capabilities, allowing other developers to download, modify, deploy, or continue training their own models based on it.

This doesn't mean the model has become cheap. K3's weight files exceed 1.5 TB, requiring multiple high-end accelerator cards just to load the complete model. Deploying it for production environments could involve significant hardware investment. Its "openness" primarily addresses whether enterprises can own the model and avoid dependence on a single API provider; it doesn't mean everyone can run it on their home computer.

Even so, open weights will still impact the business of closed-source companies.

When a sufficiently capable model can be downloaded, enterprises have the opportunity to deploy it on their own servers, keep their data internal, customize the model, and avoid perpetually paying token prices set by a particular company. The barriers that closed-source models previously built through their lead in capability will also be eroded more quickly.

Therefore, after the release of K3, the debate quickly moved beyond technical leaderboards into the realms of policy and commercial interests within the US AI industry.

Behind the open-source vs. closed-source debate

On July 24th, companies and institutions including OpenAI, Google, Microsoft, Nvidia, AMD, Meta, and Hugging Face publicly supported open-weight models, opposing the US government's rush to restrict AI models that could be downloaded and deployed.

Anthropic and Amazon were not on the signatory list. This led to external speculation that Anthropic might be using security and national security as pretexts to limit the development of open models and block competitors for its own closed-source business.

On the very day K3's weights were opened, Anthropic CEO Dario Amodei published an article responding to the controversy.

He stated that Anthropic has never advocated for a complete ban on open-weight models. He described open models that lack dangerous capabilities as a public good, lowering usage costs and creating value for businesses, developers, and researchers.

However, Amodei's core position remained unchanged.

He continued to advocate for restricting the flow of advanced chips and chip-making equipment to China, cracking down on so-called "industrial-scale model distillation," and requiring mandatory safety testing before the release of sufficiently capable models, regardless of whether they are open or closed source.

While ostensibly about safety, this debate also has a very practical business dimension underneath.

Closed-source companies want to continue controlling the strongest models, charging fees through APIs and subscription services. Open models, on the other hand, disseminate capabilities among more companies, lower prices, and shorten the distance between latecomers and leading firms. The faster the openness, the harder it is for a few companies to lock the most powerful capabilities permanently within their own servers.

K3 comes from China, is large in scale, and its capabilities are already close to the frontier. For proponents of open models, K3 proves that the open path can still produce first-class models. For Anthropic, it also signifies that the capability advantage built by US companies at enormous cost could be diffused at a faster pace.

As a result, Moonshot AI has gained a kind of capital value it previously lacked.

It is no longer just an AI application company with a large Chinese user base, nor just a foundational model startup waiting for its business model to mature. After the release of K3, it began to be revalued within the multiple competitive dynamics of global open vs. closed source, US vs. China, and model capability vs. compute control.

The $3.5 billion financing and the $35 billion valuation are the answer provided by the capital market.

Original Link