Manus is likely to pursue an IPO in Hong Kong, its $500 million financing round is nearing completion, and its valuation has doubled to $4 billion.

On September 17, Bloomberg reported that the planned $500 million financing of Manus, a Chinese AI agent startup, is about to be completed, and its valuation will double from the previous level to $4 billion (equivalent to about 26.8 billion yuan), making it the highest-valued startup in China's AI agent sector. People familiar with the matter said that negotiations are still ongoing, transaction terms are still subject to change, and the identities of new investors have not yet been disclosed.

Manus's existing investors include Tencent, HSG and ZhenFund. According to previous reports, prior to the latest round of financing, Manus's founders and existing investors bought back shares from Meta at a valuation of $2 billion, and Tencent became Manus's largest external investor after acquiring the shares initially held by Benchmark.

Market sources said that Manus may choose to list in Hong Kong in the follow-up.

The story of Manus cannot be separated from the acquisition deal with Meta that took only seven and a half months from official announcement to cancellation.

In December 2025, Meta announced the acquisition of Manus for more than $2 billion, which was the third largest acquisition in Meta's history, second only to WhatsApp and Scale AI. After the acquisition was completed, Manus founder Xiao Hong served as Vice President of Meta, and Manus's official website once marked "Manus is now part of Meta".

Within the Meta system, Manus was quickly integrated into the advertising business system, and its daily revenue increased from about $300,000 to nearly $1.5 million, and its ARR climbed from $100 million at the time of acquisition to more than $400 million.

However, in April 2026, the Office of the Working Mechanism for Foreign Investment Security Review of the National Development and Reform Commission issued an announcement, making a decision to prohibit investment in Meta's acquisition of Manus according to law, and requiring the cancellation of the acquisition transaction. This is the first publicly halted foreign acquisition in the AI sector since the Measures for the Security Review of Foreign Investment came into effect in 2021.

Canceling the acquisition means "restoring the original state", Meta withdraws its shares and Manus returns to its original state. According to a Bloomberg News report in July, before launching a new round of financing, Manus's founding team, together with its original investors Tencent, HSG and ZhenFund, bought back all shares from Meta at the original acquisition valuation of $2 billion.

Tencent took over the equity previously held by Benchmark and immediately became Manus's largest external institutional investor; Benchmark exited after achieving several times of returns. Previously, the founding party had tried to introduce new investors in the share repurchase process, but was blocked by the regulatory authority; the regulatory opinion is that this repurchase must strictly restore the equity state before the acquisition took place.

As one of the conditions for the spin-off, Manus needs to delete relevant user data generated since December 29, 2025, and set a 48-hour data deletion window from August 23 to 24, 2026, during which affected users cannot access their accounts. Meta has also set up a firewall internally, gradually terminating data sharing and revoking employee system permissions.

On September 1, 2026, Manus officially announced the resumption of independent operation, and the founding team will continue to lead the company to develop generative AI agents for global users.

Manus is developed by Butterfly Effect Company. Its founder Xiao Hong is a post-90s serial entrepreneur who graduated from Huazhong University of Science and Technology and started his business in Wuhan. In his early days, he developed the WeChat Official Account typesetting tool "Yiban" and the WeCom customer relationship management tool "Weiban", and sold them to a unicorn company. In 2024, serial entrepreneur Ji Yichao and product manager Zhang Tao joined the team and jointly developed Manus.

In March 2025, Manus was officially launched. It can dispatch different tools to solve complex problems, complete multiple tasks such as screening resumes, creating travel itineraries and analyzing stocks, and once attracted widespread attention at home and abroad. Only 8 months after its launch, Manus's ARR exceeded $100 million, with a monthly growth rate of over 20%, processing more than 147 trillion Tokens in total and creating more than 80 million virtual computers.

As of May 2026, Manus's ARR has soared to between $400 million and $500 million, nearly 4 times higher than the $100 million at the time of acquisition. Calculated at a valuation of $4 billion and an ARR of about $450 million, the corresponding price-to-sales ratio is only about 8.9 times, which is not a high multiple for an AI agent company that is still in a period of rapid growth.

Manus's technical path chooses a completely different route from that of basic model manufacturers. Co-founder Ji Yichao wrote on Manus's official blog that the team faced a key decision at the early stage of the project: to train end-to-end agents using open-source basic models, or to build agents based on the in-context learning capabilities of cutting-edge models.

Manus chose the latter, betting on "context engineering", enabling the product to deliver improvements in hours rather than weeks, and maintaining an orthogonal relationship with the underlying models. "If model progress is a rising tide, we hope Manus will be the boat, not the pillar fixed to the seabed."

This technical choice enables Manus to call cutting-edge models such as Claude and GPT, and realize autonomous planning of complex tasks and tool invocation through context engineering. But this also means that Manus does not train its own basic model from scratch, which is why it is called a "shell-wrapping" product by some industry insiders, and it is also the biggest competitive challenge it currently faces.

The AI agent track where Manus is located is experiencing a development stage where rapid growth and fierce competition coexist.

According to IDC data, the scale of China's enterprise-level AI agent market is expected to increase from 21.2 billion yuan in 2025 to 44.9 billion yuan in 2026, and is expected to exceed 332 billion yuan in 2029, with a compound annual growth rate of more than 100%. But at the same time, agent developers are facing "dimensionality reduction attacks" from basic model manufacturers. Basic models such as Kimi from Moonshot AI and Claude from Anthropic are increasingly able to efficiently handle general desktop tasks, and their efficiency is comparable to that of Manus.

This means that Manus's core competitiveness must be built on differentiated capabilities on top of basic models, that is, in-depth optimization of context engineering, multi-tool scheduling and task orchestration, rather than the model capabilities themselves. The company needs to prove that when the capabilities of basic models continue to improve, the value of its products will not be "eaten up".

In addition, a competitor that conducts similar financing to Manus, Yanyu Technology which owns the world's first AI design agent Lovart, has a valuation of $3 billion. Manus leads with a valuation of $4 billion, but the gap is not huge.

Manus's financing move is also a key step in its capitalization path.

In May this year, according to people familiar with the matter, Manus's three founders Xiao Hong, Ji Yichao and Zhang Tao once discussed raising about $1 billion in external financing to buy back the company from Meta. If the financing is successful, the next step will be to establish a joint venture with investors, and then conduct an IPO in Hong Kong. Now that the $500 million financing is about to be completed, the market generally expects Manus to choose to list in Hong Kong in the follow-up.

This financing shows that even with the rapid iteration speed of underlying basic models and continuous industry competition pushing down product pricing, investors are still willing to bet on the AI agent track, and the market's confidence in Manus is constantly increasing.

From being acquired by Meta to resuming independence, from a $2 billion valuation to a doubled $4 billion valuation, Manus has experienced the most dramatic ups and downs a startup can experience in the past year. Its ARR has soared from $100 million to $400-500 million, daily revenue has increased fivefold within the Meta system, and millions of users and enterprises around the world have adopted it. These numbers outline Manus's industrial background as a leading player in the AI agent track.

But the challenges are equally clear: Manus does not train its own basic model, and its core value is built on context engineering and tool orchestration. When basic models such as Kimi and Claude become better and better at handling general tasks, Manus's differentiated space needs to be continuously verified.

The $4 billion valuation corresponds not only to the current value of an AI application company, but also to the market's vote on whether the AI agent track can survive independently and continue to grow under the trend of basic models "swallowing everything".

This article is from the WeChat Official Account "Unicorn Insight", written by iponews Studio, and published with authorization from 36Kr.