An invitation code was once hyped to 100,000 yuan. This made Manus (Latin for "hand and brain together") a phenomenal unicorn in the global AI agent track in 2025.
But even more "crazy" than the invitation code was its next sixteen months: bought by Meta (formerly Facebook, an American internet technology company) for $2 billion, halted by a Chinese ban, and then redeemed by Tencent at the same price.
Less than three weeks after the buyback was completed, Manus stormed back onto the capital table with a valuation of $4 billion. According to Bloomberg, Manus is advancing a financing round of about $500 million, with a post-investment valuation target of $4 billion, exactly double the price at the time of Meta's acquisition.
Potential investors include IDG Capital and Boyu Capital, while existing shareholders Tencent, Sequoia China, and ZhenFund intend to follow on. If the deal lands, Manus will become China's highest-valued AI agent startup. As of press time, neither Manus nor Tencent has issued an official announcement or response regarding the above financing news.
Behind the doubled valuation, one question has never been answered: when Meta's traffic pipeline is cut off, and when Anthropic's (an American artificial intelligence company) API (application programming interface) bill eats up nearly half of revenue every month, can this company continue to draw the growth curve of ARR (annual recurring revenue) on its own?
In this regard, Guo Tao, an expert advisor to the Wuhan Municipal Commerce Bureau, told Phoenix WEEKLY Finance that after leaving Meta's traffic support, whether Manus can maintain its original growth depends on two things: first, whether the orders originally related to Meta have exclusivity clauses or binding contracts, and whether these orders can continue after the transaction is terminated; second, whether the company itself can independently acquire customers and independently deliver.
"If revenue is highly concentrated in a single source, Meta, then after the merger is terminated, the original high-speed growth curve will most likely slow significantly, and Manus will need to re-verify its commercialization and self-sustaining ability," Guo Tao said.
ARR quadrupled in half a year, mainly relying on Meta?
In the first half of this year, Manus's revenue growth was like riding a rocket.
When Meta announced the acquisition in December 2025, Manus's ARR was about $100 million. According to Caixin and The Information, by the end of June 2026, that figure had climbed to the $400 million to $500 million range. In less than a year, revenue increased 4 to 5 times, and the company's daily revenue rose from about $300,000 to nearly $1.5 million.
Where did the money come from? Multiple industry media reports say that less than two months after completing the acquisition, Meta embedded Manus into its Ads Manager (advertising management system) and opened it to more than 10 million advertisers, then pushed it into the WhatsApp Business (business version of the instant messaging app) and Instagram (photo social platform) ecosystems.
In other words, during the seven months of Manus's fastest revenue growth, the stage of operating as "part of Meta" accounted for a considerable portion.
An enterprise-level B2B client of AI agents told Phoenix WEEKLY Finance that if the traffic support poured in through Meta's pipeline is excluded, whether Manus's revenue after independence can maintain its growth curve is uncertain.
In Guo Tao's view, Manus's explosive growth occurred during the Meta acquisition window and has obvious "event-driven" characteristics; how much is endogenous growth needs to be carefully distinguished.
"A large part of the increment comes from scenario implementation within the Meta system, targeted procurement, and ecosystem traffic diversion. It belongs to order expansion brought by binding to a single giant customer, and does not equal verification of its own customer acquisition ability," he said.
The cost of being a wrapper: gross margin squeezed from both ends
The Information reported that when Manus uses Anthropic's Claude (a large language model), it needs to pay Anthropic an average of $2 for each completed task. Nvidia CEO Jensen Huang pointed out at GTC that the computing power required for agentic AI is 100 times higher than previously expected.
According to Manus's public disclosure, its gross margin is about 50%. This means that out of every $100 in revenue, about half is used to cover costs mainly consisting of computing power. CICC pointed out in a research report that the gross margin of the overseas API model is generally between 40% and 50%, basically consistent with the level disclosed by Manus.
The above-mentioned enterprise-level B2B client explained that Manus's model of calling third-party APIs limits gross margin in two ways. First, the rigid proportion of upstream model invocation costs is high; computing power and Token (word unit) procurement are variable costs, and it is difficult to continuously dilute unit costs as scale expands, unlike foundation model companies. Second, product pricing is constrained by upstream vendors' price adjustments, and it has no bargaining power over the underlying model.
In its September 2026 Agent Economy Report, CICC proposed a directly quotable criterion: 1 yuan of Token cost must bring back 10 yuan of revenue, otherwise it is just a model wrapper.
In Guo Tao's view, the barriers of wrapper application companies mainly lie in Prompt engineering, agent workflows, and scenario delivery, which are relatively thin, and the capital market will therefore discount them. "Unless a company forms an extremely strong barrier in a vertical industry can it hedge against dependence on upstream."
However, the above-mentioned enterprise-level B2B client also reminded that Manus's core value lies in product-layer workflow orchestration and user experience, which self-developed model companies may not be good at. Its valuation premium should rest on "product definition capability," not technical barriers.
Why was the $2 billion deal halted?
Manus's story cannot avoid a ban.
On December 30, 2025, Meta announced the acquisition of Manus's parent company Butterfly Effect Pte. Ltd. (Butterfly Effect Technology Co., Ltd., registered in Singapore), with a merger amount exceeding $2 billion, the third-largest merger in Meta's history since its founding.
On April 27, 2026, according to the website of the National Development and Reform Commission, the Office of the Working Mechanism for Foreign Investment Security Review (National Development and Reform Commission) made a decision prohibiting investment in the foreign acquisition of the Manus project in accordance with laws and regulations, requiring the parties to revoke the acquisition transaction.
According to Guangming Online, this is the first publicly halted foreign acquisition in the AI field since the implementation of the Measures for the Security Review of Foreign Investment in 2021, and also the strictest review conclusion under the framework of that measure: the "prohibition of investment" tier.
"The essence of this case is that the path of changing a shell and selling it abroad no longer works," an enterprise strategy consultant focused on AI Agent commercialization told Phoenix WEEKLY Finance. The Manus team, data, and R&D were originally all in China, but later the headquarters was moved to Singapore, the Chinese subsidiary was divested, and data was transferred overseas, then sold to Meta, touching two red lines: outflow of core technology and outbound data transfer.
"For other AI companies with foreign backgrounds, the model of 'changing a vest and going overseas' to sell to foreign capital is no longer viable," he said.
Guo Tao pointed out that for AI startups with foreign backgrounds, exit methods are changing. "Entrepreneurs and investors need to plan early and think clearly at the financing stage. Don't treat 'selling to a big American company' as the only hope."
He reminded that AI is a key field, and foreign capital seeking acquisitions must pass the security review. Going overseas does not mean bypassing national security assessment. How the equity structure is set up and how overseas shareholders are arranged will become increasingly important.
Tencent steps in, as "lead party" rather than "takeover party"?
According to an exclusive report by Caijing on September 1, Tencent led the push for Manus's buyback, becoming the largest external shareholder but not controlling it; the actual controller remains the Manus founding team. Sequoia China and ZhenFund each hold about 10%. The agreement was launched in early July and approved and signed at the end of August.
The transaction consideration was $2 billion, flat with Meta's previous acquisition price. In the past eight months, Manus's ARR has climbed from $100 million to $400 million to $500 million.
"Tencent acting as lead party rather than takeover party is more like a strategic investment, not strategic integration," the above-mentioned enterprise strategy consultant said. This is Tencent's consistent approach of "investing in the ecosystem, not controlling," giving money and resources and letting the company run itself.
In Guo Tao's view, Tencent is currently a financial investor, and as long as it does not obtain major voting rights, it will not interfere in product R&D and business decisions. "But the capital market will watch to see whether it has a veto in major decisions. At this stage, it will not shake Manus's independence."
He pointed out that Manus is a key layout for Tencent in the general Agent track. Independent operation is clearer in terms of compliance.
Hong Kong stock listing: path is clear, obstacles remain
After the buyback was completed, Manus's listing path has shifted from "selling to Meta" to "Hong Kong IPO." According to Caijing, the company is restructuring its architecture and building a domestic joint venture entity to meet the compliance requirements for a Hong Kong listing.
But obstacles on this path remain. "Whether outbound data transfer is compliant is the primary issue," the above-mentioned enterprise-level B2B client said. Manus's product training has called on a large amount of domestic data, but it has not fulfilled statutory procedures such as security assessment, certification, or standard contract filing for outbound data transfer, which regulators have determined touches a red line. At the same time, building a domestic joint venture entity as the listing entity also involves complex cross-border equity sorting.
According to Caijing, Manus has deleted user data generated after December 29, 2025, starting from August 23, 2026. That date was the day Meta officially announced the acquisition and is regarded by the outside world as the dividing point for the transfer of corporate control.
“AI regulation iterates quickly, and data compliance flaws may trigger a suspension of the listing process.” Guo Tao stated bluntly that Manus's Hong Kong stock listing will face three obstacles: first, whether historical data outbound transfers have completed the legally required security assessment; second, the prospectus must fully disclose data sources, storage locations, and cross-border flow paths, and non-compliant operations will find it very difficult to pass inquiries; third, the foreign investment security review event itself will be a key focus of the exchange.
(Image source: Manus WeChat service account)
This article is from the WeChat public account “Phoenix WEEKLY Finance”, author: Xu Mengyi
