Meta Manus AI Deal Blocked: What Happened and Why China Stopped It
Key Takeaways
- This is not a new deal. Meta agreed to buy the AI agent company Manus in December 2025, for a reported price of more than 2 billion dollars. China blocked the deal in April 2026 and ordered it unwound. The recent rise in searches is renewed interest in that blocked deal, not news of a fresh acquisition.
- Manus is independent again. It formally resumed independent operations on 1 September 2026, under its founding team, after its earlier investors bought it back from Meta.
- The blocking body was an office inside China’s National Development and Reform Commission, the NDRC, acting under China’s foreign investment security review system. Its public notice was brief and did not name Meta, name a specific legal provision, or give a detailed reason.
- The unwind had a real cost for users. Data created on or after 29 December 2025 was deleted for affected accounts in late August 2026, after a short backup window. Manus described the deletion as part of its separation from Meta.
- Manus has since raised more than 500 million dollars as an independent company, in a round announced on 8 October 2026, and shipped Manus 2.0 on 28 September 2026. Meta, meanwhile, launched its own personal AI agent, Muse, in September 2026.
- The reported price of more than 2 billion dollars was never confirmed by Meta or Manus. Treat it as a reported figure, not an official one.
Read this first: there is no new Meta-Manus deal
If you arrived here because a headline or a search result suggested that Meta has just bought Manus, stop and read this paragraph first, because the headline is out of date. Meta did agree to buy Manus. That happened on 29 December 2025. Four months later, on 27 April 2026, China blocked the deal and ordered the two sides to withdraw it. By September 2026 Manus was operating independently again, and by October 2026 it had raised fresh money as an independent company.
So why is the phrase “meta manus ai deal” rising in searches again in November 2026, by about 300 percent according to Google Trends data for the past seven days? There is no single verified trigger in the reporting we reviewed. The most likely explanation is simpler and less dramatic: an old, unusual story is being rediscovered. A deal in which one of the world’s largest technology companies bought a fast-growing AI startup, only for a government to order the completed purchase reversed months later, is the kind of story that keeps circulating, especially as Manus launches new products and Meta pushes its own rival agent. Renewed interest is not the same as a new event, and this article will keep the two apart throughout.
That distinction matters for anyone thinking about using Manus, investing in the AI agent market, or simply trying to understand how AI deals now work across the United States and China. The Meta-Manus story is not a story about a takeover. It is a story about a takeover that failed, about what happens when a deal closes and is then ordered undone, and about how quickly a startup can be sold, separated, bought back and refinanced in the space of ten months.
What Manus is, and why it mattered
Manus is a general-purpose AI agent, built by a company called Butterfly Effect. It was founded in Beijing in 2022 and later based itself in Singapore. Unlike a chatbot that answers a question and stops, Manus was designed to carry out multi-step tasks on its own: researching a topic across many sources, building a spreadsheet, writing code, booking travel, or producing a working website from a plain-language request, inside its own virtual computer. It launched publicly with a viral demonstration in March 2025, and that demonstration did a great deal of the company’s early marketing for it.
The timing mattered. In early 2025 the AI industry was shifting from chatbots that answer questions to agents that do things. Manus arrived at exactly that moment, with a product that looked like the shift made concrete. By December 2025 the company said it had crossed 100 million dollars in annualised recurring revenue, eight months after launch. That figure came from the company itself, in a post in December 2025, and it is the last revenue figure Manus itself has confirmed. Later reports put its annualised revenue higher by mid-2026, in the region of 400 to 500 million dollars, but those figures come from third-party reporting, not from the company, and this article does not treat them as confirmed.
Manus also moved its people before the Meta deal. The founding team was Chinese, and in mid-2025 the company moved its staff to Singapore. That move turns out to be central to the story, because it did not, in the end, insulate the deal from Chinese review. A company can move its headquarters and its staff. It cannot so easily move its origins, its founders, or the regulatory interest of the country where it was founded.
Why Meta wanted Manus
Meta’s interest in Manus is easy to understand, even without access to the internal reasoning. Meta was, and is, trying to build AI products that do things for people rather than merely answering them. Manus offered a working agent product, a team of roughly 100 employees who knew how to build one, and a brand that had already captured public attention. Buying Manus would have given Meta a shortcut: a proven agent, a proven team, and a head start in a market where every large company was racing to ship agents of its own.
The deal was announced on 29 December 2025. Meta said Manus would become part of its consumer and business products. The price was not disclosed by either side. Reporting at the time, including from TechCrunch, put the value at more than 2 billion dollars, with some later reports putting it as high as 2 to 3 billion dollars. A later review of the Manus deal and its aftermath gives the same reported range and dates the blocking to 27 April 2026. At the company’s own stated revenue figure of 100 million dollars in annualised recurring revenue, a price above 2 billion dollars would have been roughly twenty times that revenue run rate, which tells you how highly Meta valued the team and the technology rather than the current sales alone.
Meta also set a condition that showed it understood the political sensitivity. A Meta spokesperson said there would be no continuing Chinese ownership interest in Manus after the deal closed, and that Manus would wind down its services and operations in China. In other words, Meta tried to present the deal as a clean break: a Singapore-based company, bought by an American company, with its Chinese links cut. As events showed, that was not enough.
For readers who want the wider context on what agents are and why large companies are buying them, our explainer on AI agents, what they are and how they work covers the technology in detail. The Manus deal is best understood as one of the most expensive bets yet on that technology.
The timeline: from announcement to unwind
The story moves quickly, and the dates are the best way to hold it together.
In March 2025 Manus launched publicly with its viral agent demonstration. In mid-2025 it moved its staff to Singapore, after winning backing from the United States venture firm Benchmark, which had led a funding round reported at 75 million dollars in April 2025. In December 2025 Manus said it had crossed 100 million dollars in annualised recurring revenue.
On 29 December 2025 Meta and Manus announced the deal. Manus said it was joining Meta. The deal closed the same day, according to later reporting. That speed matters: this was not a long, drawn-out acquisition with months of closing conditions. It was announced and closed, and Meta began integrating the Manus team and technology.
On 8 January 2026 a spokesperson for China’s Ministry of Commerce said Beijing would investigate the deal, to check whether it breached export controls or foreign investment laws. That was the first public sign that the deal might not survive, and it came barely ten days after the announcement.
On 27 April 2026, four months after the deal was announced, the Foreign Investment Security Review Office, part of the NDRC, published a short notice. It said the commission had decided to prohibit foreign investment in the Manus project in accordance with laws and regulations, and required the parties to withdraw the acquisition transaction. The notice was, by several accounts, a single sentence long. It did not name Meta. It did not name a specific legal provision. It did not give a detailed reason.
That brevity is itself part of the story. Lawyers who follow China’s foreign investment review system described it as the first publicly disclosed case of an AI-sector deal blocked under the relevant rules, Order No. 37, the Foreign Investment Security Review Measures issued jointly by the NDRC and China’s Ministry of Commerce in 2021, since those measures took effect. The order did not ask Meta to renegotiate the deal or to accept conditions. Article 12 of the measures allows an outright prohibition of a transaction judged to touch national security, and the decision used that power.
In the weeks around the blocking, the two Manus co-founders, Xiao Hong and Ji Yichao, were reportedly barred from leaving China. That report comes from secondary reporting and has not been confirmed by an official source we could verify, so treat it as reported rather than established. It is included here because it appears in several accounts of the period, and because it helps explain the pressure the company was under, but it should not be stated as a confirmed fact.
In July 2026, according to reporting from Caixin and others, Tencent, ZhenFund and HSG bought Manus back from Meta at a reported valuation of about 2 billion dollars, the same level as the original deal. Tencent became the largest shareholder, while remaining a minority holder, according to those reports. The exact ownership percentages have not been published, and this article does not state them as fact.
On 11 August 2026 Manus told users about the separation. It said it would delete data created from 29 December 2025 for users in certain jurisdictions, after a backup window ending on 22 August 2026. Affected accounts had from 11 August until 23 August to run a backup tool, with free service during the window. Data was then deleted over 23 to 24 August, and a restore portal opened on 25 August. Manus said the data removal was part of its separation from Meta. Manus has not published how many accounts were affected or how much data was deleted, and this article does not guess at either figure.
On 1 September 2026 Manus said it had formally resumed independent operations, with its founding team continuing to lead the company. On 28 September 2026 it launched Manus 2.0, built on a new in-house execution system called Cascade, alongside Manus Studio, a cloud computer, automations, and Cue, a separate app that gives personal agents their own digital identities.
On 8 October 2026 Butterfly Effect, Manus’s parent company, announced that it had raised more than 500 million dollars, its first funding round since the unwind. A timeline of the deal from announcement to independence records the same sequence: the December 2025 announcement, the April 2026 prohibition, and the return to independence in September 2026. Boyu Capital and IDG Capital led the round, with Tencent, HSG and ZhenFund also taking part. The company did not disclose its valuation after the round. Bloomberg had reported in September that the round could value Manus at about 4 billion dollars, but that remains a report, not a company statement, and it should be read as such.
Why China blocked it
The short answer is that we do not know the full reasoning, because the Chinese authorities did not publish it. The longer answer is that several factors, all visible in the public record, point in the same direction.
First, the legal instrument was the foreign investment security review, a system designed to let China block foreign investment in projects that touch national security. The NDRC office that issued the decision is the body that runs that review. The decision was framed in terms of laws and regulations, without naming a specific provision, which is consistent with a national security review rather than an ordinary commercial or competition decision.
Second, the technology mattered. Manus was not a generic software company. It was an AI agent company, building systems that can carry out tasks autonomously on a computer. AI agents are now treated by governments as a strategically important technology, and a deal that would have transferred a leading Chinese-founded agent company, its team and its technology into American ownership was always likely to attract scrutiny.
Third, the relocation did not settle the question. Manus had moved its staff to Singapore and was headquartered there, and Meta had said there would be no continuing Chinese ownership. But the company was founded in China, its founders were Chinese, and it had Chinese investors. The NDRC’s decision showed that, in the view of the Chinese authorities, those origins still brought the deal within the scope of the review. That is a significant point for any AI company with a similar history: moving offshore does not necessarily move you outside the reach of the country where you started.
Fourth, export controls were in the background from the start. The Ministry of Commerce said in January 2026 that it would investigate whether the deal breached export controls or foreign investment laws. Whether export controls were the decisive factor, or simply one factor among others, has not been confirmed. The NDRC notice itself did not say.
It is also worth noting what the decision was not. It was not a competition ruling, and it was not a ruling that Manus had done anything wrong. Nothing in the public record suggests that Manus or Meta broke the law in announcing the deal. The decision was a security review outcome: a judgement that the transaction should not proceed, followed by an order to reverse a transaction that had already closed.
That last point, the reversal of a closed deal, is what makes this case unusual. Most blocked deals are blocked before they close. Here the deal had closed, integration had begun, and the parties were then told to undo it. Unwinding a closed deal means separating teams, systems and data that have already been joined, and that is a far more disruptive exercise than simply walking away from an unsigned agreement. The data deletion in August 2026 was one visible consequence of that exercise.
For the wider picture on how the United States and China are competing in AI, and how that competition shapes deals like this one, see our report on US and China AI competition in 2026, and who is really winning.
What the unwind actually meant in practice
An order to “withdraw the acquisition transaction” sounds abstract. In practice it meant several concrete things, and they happened in a specific sequence.
Meta cut Manus staff off from its internal data systems. Meta employees were barred from using Manus tools. Manus announced that it would operate as an independent company again. The earlier investors bought the company back. And users were told that data created during the Meta period would be deleted.
That data deletion deserves attention, because it is the part of the story that affected ordinary users most directly. If you used Manus between 29 December 2025 and August 2026, and your account was in one of the affected jurisdictions, your data from that period was deleted unless you backed it up during the window Manus provided. Manus gave users from 11 August to 23 August to back up, offered free service during that window, and opened a restore portal on 25 August. That is a short window for a deletion of that scale, and it is one reason the unwind is remembered as painful rather than merely procedural. Manus has not said how many accounts were affected, and this article will not estimate a number.
There were also consequences for Meta. The company had paid, or agreed to pay, a reported sum of more than 2 billion dollars, integrated a team of about 100 people, and then had to hand the company back. The buyback, at a reported valuation of about 2 billion dollars, suggests Meta did not lose the full amount in a simple sense, but the time, the integration effort and the strategic opportunity were lost. Meta also lost the head start that Manus would have given it in agents.
The episode also raised a question that has not been fully answered: what happens to the work done during the Meta period? Manus said it would delete data created from 29 December 2025 for users in certain jurisdictions. What happened to the technology, the code and the product development from that period is less clear in the public record, and this article does not claim to know. That is a gap, not a detail to be filled in by speculation.
Manus after independence: the comeback
Manus did not disappear after the unwind. In fact, its recovery has been fast by any measure.
The company resumed independent operations on 1 September 2026, with its founding team intact. Later that month it shipped Manus 2.0, a substantial product release built on a new execution system called Cascade, with Manus Studio, a cloud computer, automations and Cue, the personal agent app. The message was clear: Manus was not a wounded company limping back to independence. It was a company with a new product and a new roadmap, focused on agents embedded in workflows and proactive agents rather than one-off task runs.
Then came the money. On 8 October 2026 Butterfly Effect announced a funding round of more than 500 million dollars, led by Boyu Capital and IDG Capital, with Tencent, HSG and ZhenFund taking part. Analysts quoted in the reporting said the raise showed that the short-term fallout from the Meta case had been contained and that investors were willing to back Manus as an independent company. That is a plausible reading, but it is an analyst judgement, not a fact, and it should be read as such.
The valuation was not disclosed. Bloomberg had reported in September that the round could value Manus at about 4 billion dollars, double the reported price of the Meta deal. That figure has been widely repeated, but it remains a report. Manus itself has not confirmed it. Readers should therefore treat the 4 billion dollar figure as reported, not established.
There are also reports that Manus is considering a Hong Kong listing, with Reuters reporting that an initial public offering would not begin until at least 2027. That too is a report, not a company announcement, and the timing and the venue should be treated as unconfirmed.
What is confirmed is that Manus is operating, shipping and raising money as an independent company, less than a year after being bought by Meta. That is a remarkable recovery, and it is one reason the story keeps being rediscovered. It is also a reminder that a blocked deal is not necessarily the end of the company involved. It can be, for the company, a painful interruption followed by a new start.
What Meta did next
Meta did not stop building agents after the Manus deal was unwound. In September 2026 it launched Muse, its own personal AI agent. That launch is relevant because it shows Meta’s response to losing Manus: build the capability in-house rather than buy it. Muse is now, in effect, Meta’s answer to the product it tried to acquire.
Whether Muse can do what Manus does, and whether it will succeed where the acquisition failed, is a separate question, and one this article does not try to answer. The products are different, the teams are different, and it is too early to judge. What can be said is that Meta’s agent strategy did not end with the unwind. It changed shape.
The wider context is also worth noting. The Manus deal was one of several large AI deals in 2025 and 2026 in which a large company tried to buy, hire or otherwise absorb a smaller AI team. Some of those deals succeeded. Some were restructured as hiring deals or licensing deals rather than outright acquisitions, precisely to avoid regulatory scrutiny. The Manus case is the most dramatic example of what happens when a deal goes ahead anyway and is then blocked. It is a cautionary tale for any large company considering a similar acquisition, and for any startup considering a similar sale.
For the financial side of the AI industry, including how large AI companies are valued and funded, see our report on Anthropic’s IPO, oversubscribed and valued in the trillions.
What this means for Manus users
If you use Manus, or are thinking about using it, the blocked deal raises practical questions, and the answers are mixed.
On the positive side, Manus is independent, shipping and funded. Manus 2.0 and Cue are real products, launched after independence, and the October 2026 funding round suggests investors are confident in the company’s future. The founding team is still in charge, which provides continuity.
On the cautionary side, the ownership picture has changed more than once in a short period, and the full cap table has not been published. Tencent is reported to be the largest shareholder, while remaining a minority holder, but the exact stake is not public. That matters for users who care about where their data is held and who ultimately controls the company. It is not a reason to avoid Manus, but it is a reason to read the company’s privacy and data policies carefully, and to keep your own backups of important work.
The data deletion in August 2026 is also a lesson. If you rely on an AI agent for important work, keep your own copies of that work outside the agent’s system. A corporate change, a regulatory order or a product change can lead to data being deleted, and a short backup window may be all you get. That is not specific to Manus. It is good practice for any cloud service, and the Manus case shows why.
Finally, do not assume that Manus is a Meta product. It is not. If you see a headline, a social media post or a search result that says Meta bought Manus, check the date. If it is from December 2025, it is describing the announcement. If it is presented as current news in November 2026, it is out of date. Manus is independent, and Meta is building its own agent, Muse.
What this means for AI dealmaking
The Manus case has implications that go beyond the two companies involved.
First, it shows that AI deals are now subject to national security review in a way they were not a few years ago. The NDRC’s decision was described as the first publicly disclosed AI-sector block under Order No. 37. That suggests the review system is now being applied to AI, and that future deals will be assessed in the same way.
Second, it shows that a company’s history matters as much as its current headquarters. Manus had moved to Singapore, and Meta had promised to cut its Chinese links, but the deal was still blocked. Founders, investors and origins all count. A startup that wants to be acquired by a foreign buyer needs to think about its full history, not just its current address.
Third, it shows that closing a deal quickly does not protect it. The Meta-Manus deal closed on the day it was announced, and it was still reversed four months later. Speed is not a substitute for regulatory clearance, and a deal that closes before review is complete carries the risk of being unwound.
Fourth, it shows that unwinds are messy. Separating integrated teams, systems and data is disruptive, and users can bear part of the cost, as Manus users did when their data was deleted. Any company considering a deal that might be reviewed should think about what an unwind would look like in practice, not just in theory.
None of this means AI deals will stop. It means they will be structured more carefully, with more attention to regulatory risk, and that some deals will be restructured as hiring or licensing arrangements rather than outright acquisitions. That shift was already under way before the Manus case. The Manus case is likely to accelerate it.
Why the story is trending again
Google Trends data for the past seven days shows searches for “meta acquires manus ai” up about 300 percent. That is a substantial rise, but it is not evidence of a new event. Search interest can rise for many reasons: a product launch, a news article, a social media post, or simply the rediscovery of an old story by a new audience.
In this case, there are several plausible contributors. Manus launched Manus 2.0 on 28 September 2026 and announced a large funding round on 8 October 2026, both of which brought the company back into the news. Meta launched Muse in September 2026, which invited comparisons with the agent it had tried to buy. And the sheer unusualness of the story, a 2 billion dollar deal ordered undone, makes it the kind of story that is shared and re-shared.
What we did not find, in the reporting reviewed for this article, is a new announcement, a new filing, or a new statement from Meta, Manus or the Chinese authorities that would explain the rise as a response to fresh news. If such a development exists, it has not been verified here, and readers should treat any claim of a new deal with caution until it is confirmed by a primary source.
That is the most important practical advice in this article: check the date on any Manus-Meta headline. If it is from December 2025, it is the announcement. If it is from April 2026, it is the blocking. If it is presented as new in November 2026, ask what is actually new, and look for a primary source before believing it.
What to watch next
Several things are worth watching as the Manus story continues.
The first is Manus itself. Will Manus 2.0 and Cue find a lasting audience? Will the company publish more about its ownership, its data policies and its revenue? Will the reported Hong Kong listing happen, and if so, when? Those are the questions that will determine whether the recovery is lasting.
The second is Meta. Will Muse develop into a serious rival to Manus and other agents? Will Meta try another acquisition in the agent space, and if so, how will it structure the deal to avoid the fate of the Manus purchase? The answers will tell us how much Meta learned from the experience.
The third is the regulatory picture. Will China block other AI deals under the same review system? Will other countries apply similar reviews to AI acquisitions? The Manus case was the first publicly disclosed AI block under Order No. 37. It is unlikely to be the last, and the next case will show whether the Manus decision was a one-off or the start of a pattern.
The fourth is the broader United States-China AI relationship. The Manus deal was blocked at a time of intense competition and tension between the two countries over AI. How that relationship develops will shape not only future deals, but the ability of AI companies to operate across borders at all.
What We Could Not Verify
- The exact price Meta agreed to pay for Manus. Reporting puts it at more than 2 billion dollars, with some reports suggesting 2 to 3 billion dollars. Neither Meta nor Manus disclosed a price, and this article treats the figure as reported, not confirmed.
- The full reasoning behind the NDRC’s decision. The public notice was brief and did not name a specific legal provision or give a detailed reason. Secondary reporting attributes the decision to national security grounds tied to outbound investment and technology export rules, but that attribution has not been confirmed by an official source.
- Whether the two Manus co-founders were barred from leaving China. That has been reported in secondary sources but not confirmed by an official source we could verify.
- The exact ownership of Manus after the buyback. Reporting says Tencent, ZhenFund and HSG bought the company back at a reported valuation of about 2 billion dollars, with Tencent as the largest but still minority shareholder. The percentages have not been published, and the cap table has not been disclosed.
- The reported valuation of about 4 billion dollars for the October 2026 funding round. Bloomberg reported that figure in September 2026, but Manus did not disclose a valuation, and it should be treated as a report, not a company statement.
- Manus’s annualised revenue after December 2025. The company confirmed 100 million dollars in December 2025. Later reports put the figure at 400 to 500 million dollars by mid-2026, but those figures come from third parties and have not been confirmed by the company.
- How many Manus user accounts were affected by the August 2026 data deletion, and how much data was deleted. Manus has not published those figures.
- What, if anything, is driving the current rise in searches for the Meta-Manus deal. No new announcement or filing was found in the reporting reviewed for this article, and the rise is described here as renewed interest, not as a response to a verified new event.
- Whether Manus will list in Hong Kong, and when. Reuters reported that a listing would not begin until at least 2027, but that is a report, not a company announcement.
The bottom line
The Meta Manus AI deal is a story about a purchase that happened and was then undone. Meta agreed to buy Manus in December 2025 for a reported price of more than 2 billion dollars. China blocked the deal in April 2026 and ordered it unwound. Manus bought itself back, resumed independent operations in September 2026, shipped Manus 2.0, and raised more than 500 million dollars in October 2026. Meta built its own agent instead.
If you remember one thing from this article, make it this: Manus is not a Meta product. It was, briefly, a Meta acquisition. It is now independent. Any headline that says otherwise is either old or wrong, and the date on the headline will tell you which.