Manus AI Agent Moves Back Toward Independence as Meta Deal Unwinds

The Manus AI agent is moving back toward independence after Beijing forced Meta to reverse an acquisition that had already closed. Early Chinese investors reportedly plan to repurchase the startup, while Meta has separated operations and stopped sharing data.

That creates a rare reversal. Manus joined Meta expecting greater resources and distribution, but its future now depends on rebuilding an independent ownership structure. The transaction has shifted from integration to separation within months.

The dispute also reaches beyond one startup. Meta wanted Manus as part of its push into software that completes multistep tasks, not just chatbot conversations. Beijing treated the Chinese-founded company as a strategic technology asset despite its relocation to Singapore.

The central question is no longer whether Meta can integrate Manus. It is whether Manus can preserve its product momentum while its previous investors reconstruct the business outside Meta.

The Manus Meta Deal Is Moving From Acquisition to Separation

The reported buyback would complete a reversal that Meta has already started operationally.

Early Chinese backers HSG, ZhenFund, and Tencent plan to buy Manus back from Meta, according to a reported buyback plan published by Reuters. The report originally came from The Information, which cited two people with direct knowledge.

Reuters said it could not independently verify the plan. Meta, Manus, and the named investment firms did not immediately comment. That verification gap matters because participation in discussions does not guarantee financing, regulatory approval, or a completed transfer.

The investors reportedly want to acquire Meta’s position at the same valuation used in the original transaction. HSG and ZhenFund were considering fresh capital for the purchase, while Benchmark was not expected to participate.

The proposed buyer group is important. HSG, ZhenFund, and Tencent backed Manus before the Meta transaction, so the plan resembles a restoration of earlier ownership more than a conventional sale.

The transaction would still leave significant questions about control. Tencent’s participation could provide capital and strategic support, but Manus reportedly expects to remain independent rather than become another Tencent operating unit.

The ownership arrangement has not been publicly finalized. It remains unclear how much each investor would contribute, whether Manus management would receive additional equity, or which entity would hold the intellectual property.

Meta, meanwhile, has moved beyond preliminary discussions about separation. A systems split reported by Bloomberg established a practical firewall between the companies.

Meta reportedly barred Manus employees from accessing its internal data systems. Meta employees also lost permission to use Manus tools for internal projects. Data sharing between the companies stopped.

Those measures change the Manus Meta deal in concrete ways. Manus can no longer operate as an integrated Meta asset while the ownership question remains unresolved.

They also reduce the risk of additional technology or information moving between the businesses during the unwind. That distinction matters because Beijing’s intervention concerned more than the corporate registration of Manus.

The dispute involves control over technology, employees, data, and cross-border investment. An operational split addresses those areas before the financial transfer is complete.

For users, the separation does not automatically mean that the Manus service is shutting down. Manus continued releasing integrations after Meta began dismantling the relationship.

However, continued product availability should not be confused with settled ownership. The company is operating through a transition whose financing, governance, and jurisdiction remain under negotiation.

Beijing’s Order Reached Beyond Manus’s Singapore Address

Manus discovered that moving operations offshore did not erase the regulatory importance of its Chinese origins.

Meta announced the acquisition in December 2025. Manus was based in Singapore, and Meta said most of its employees worked there.

The company nevertheless traced its development to Beijing-registered entities. Its earlier investors included prominent Chinese venture firms and Tencent, one of China’s largest technology companies.

China opened a review in January 2026. The Commerce Ministry said businesses involved in outward investment, technology exports, data transfers, and cross-border acquisitions must follow Chinese law.

In April, the National Development and Reform Commission prohibited the foreign acquisition of Manus. Its foreign investment security review office required the parties to withdraw from the transaction.

The short public order offered little explanation. The acquisition block nevertheless established that Beijing considered its authority applicable despite Manus’s Singapore base.

Meta responded that the transaction had complied with applicable law. It also said it expected an appropriate resolution to the inquiry.

That response reflected the legal conflict at the center of the case. Meta had completed a transaction involving a Singapore company, while Chinese authorities focused on the startup’s earlier corporate, technological, and investment connections.

The original acquisition was unusual because a major American platform was buying a successful AI company with strong Chinese roots. It occurred while both governments were increasing controls around advanced technology and cross-border capital.

China’s intervention suggests that corporate relocation alone does not decide which government can claim an interest. Regulators can examine where technology originated, who developed it, which entities funded it, and how data previously crossed borders.

That logic places pressure on other internationally structured AI startups. A company can move its headquarters, employees, and commercial operations without fully removing its technology from its original regulatory history.

Omdia analyst Lian Jye Su told the Associated Press that Beijing was treating AI talent and capabilities as national security assets. He said the action could shape future acquisitions involving Chinese deep-technology companies.

The order also mirrors a broader pattern of technology controls. The United States has used export restrictions, investment reviews, and entity lists to limit certain technology transfers to China.

Beijing is now demonstrating its own ability to prevent strategically valuable technology from moving into American ownership. Manus became a visible test because the deal had already closed.

This timing makes the case more consequential than a rejected acquisition proposal. Regulators did not merely prevent future integration. They compelled two companies to reverse integration steps that had already begun.

That introduces a new risk for cross-border buyers. Closing a transaction does not necessarily end regulatory uncertainty when technology, data, or corporate lineage connects several jurisdictions.

For founders, the lesson is equally direct. Relocation can expand market access, but it does not guarantee that earlier regulators will treat the company as fully detached.

Why the Manus AI Agent Still Attracts Its Former Backers

The investors are reportedly returning because Manus became more commercially valuable while the Meta transaction was unraveling.

A May 2026 Manus product example shows the agent producing a reviewable plan before building a Shopify storefront. This is evidence of continued product operation during the ownership unwind, not evidence that the buyback is complete. Source: Manus official blog.

Manus develops a general-purpose agent, meaning software that plans and performs multistep computer tasks with limited supervision. Reported use cases include coding applications, conducting market research, and preparing business budgets.

That task-oriented design distinguished Manus from chatbots that primarily generate text in response to individual prompts. An agent must maintain a goal, select tools, manage intermediate steps, and return a completed result.

Manus reported passing $100 million in annual recurring revenue before the acquisition. Annual recurring revenue is a standardized estimate of subscription revenue over one year.

The later buyback report said its annualized revenue run rate had increased to between $400 million and $500 million. Reuters attributed that estimate to The Information and did not independently verify it.

If accurate, that growth changes the economics of the reversal. The investors would be attempting to reacquire a company with substantially greater reported revenue at the original transaction valuation.

It also explains why Manus has options despite losing Meta’s direct support. A product generating material subscription revenue can finance operations, attract outside investors, and negotiate from a stronger position.

However, revenue run rate is not the same as audited annual revenue. It extrapolates recent performance and can change quickly when a subscription product is growing, discounting access, or experiencing unusual demand.

The distinction is especially important for AI services. High usage can increase both revenue and inference costs, which are the computing expenses created when a model processes user requests.

Neither Reuters nor Manus disclosed profitability, customer retention, compute margins, or the share of revenue coming from short-term consumer subscriptions. Those missing numbers determine whether the reported growth represents a durable business.

Still, product activity continued after the separation began. Manus released integrations involving Similarweb and Shopify, indicating that its team could keep shipping outside Meta’s internal systems.

Those integrations support a practical agent strategy. Rather than competing only on an underlying language model, Manus can connect models with business data and applications where users already work.

That approach puts Manus in competition with a growing group of agent products. OpenAI is expanding ChatGPT beyond conversation into research and task execution. Google is connecting Gemini with productivity tools and search.

Anthropic has focused heavily on coding agents and computer interaction. Smaller companies are also building browser operators, software-development agents, and task-specific automation systems.

Manus does not need to train the largest foundation model to compete in this market. It needs to make several models, tools, and interfaces work together reliably.

That product layer can be valuable because enterprises rarely judge an agent by benchmark scores alone. They care whether it completes a workflow accurately, preserves permissions, records its actions, and recovers from errors.

The Manus AI agent therefore represents something both Meta and Tencent want: a tested interface between general AI models and real user tasks.

Meta wanted that layer for its enormous consumer platforms. Tencent can see a similar opportunity across messaging, payments, cloud services, and business applications.

Independence would offer Manus a third route. It could sell across platforms without being tied to one large technology company’s distribution or product priorities.

That freedom has costs. An independent Manus would need its own compute agreements, security operations, enterprise sales, and long-term financing.

The former backers are not simply trying to restore the past. They would be buying a larger company whose operational requirements have grown alongside its revenue.

Independence Is Not the Same as Autonomy

A successful buyback can restore separate ownership without freeing Manus from strategic dependencies.

Manus’s official Shopify workflow separates agent execution from the user-controlled step of claiming the store. It illustrates why product autonomy, platform dependencies, and legal ownership are different questions. Source: Manus official blog, May 2026.

The word “independent” describes who controls the company. It does not explain where Manus gets models, computing capacity, distribution, or capital.

An AI agent depends on several layers outside its own application. These include foundation models, cloud infrastructure, browser environments, third-party APIs, and the external services where tasks occur.

If one provider changes its policies or technical interface, an agent can lose access to an important capability. That risk exists regardless of who owns the agent company.

Manus must also rebuild resources that Meta expected to provide. When the acquisition was announced, Manus CEO Xiao Hong said Meta would offer a stronger and more sustainable foundation without changing how Manus worked.

The forced separation tests that claim in reverse. Manus now has to preserve the product while removing the infrastructure, internal access, and organizational relationships associated with Meta.

Tencent’s role complicates the independence narrative. Tencent has capital, cloud infrastructure, AI models, and distribution through its applications.

Those assets could help Manus replace some benefits lost with Meta. They could also create a new concentration of influence if Tencent becomes the dominant investor or commercial partner.

The reported plan says Manus expects to remain independent from Tencent. No public ownership terms yet show how that independence would be protected.

Governance details will matter more than the buyer list. Voting rights, board seats, financing conditions, intellectual-property ownership, and exclusive commercial agreements can shape control without a full acquisition.

A proposed Chinese joint venture adds another layer. Reuters reported that Manus was considering restructuring as a joint venture incorporated in China, potentially preparing for a Hong Kong listing.

Such a structure could satisfy Beijing by bringing ownership and technology under a framework it accepts. It might also complicate Manus’s relationships with customers in other markets.

Enterprise buyers will ask where data is stored, which legal entity processes it, and whether information can move between regional systems. They will also want contractual answers about model training and government access.

The operational firewall with Meta addresses only one side of those concerns. It does not establish the final architecture of an independent Manus.

User confidence also depends on continuity. Customers need to know whether saved projects, task histories, integrations, and account permissions will survive the ownership transition.

Unverified community posts have circulated claims about data deletion and account changes. Those claims should not be treated as confirmed without a direct company notice or enforceable policy document.

The more reliable fact is narrower. Meta and Manus reportedly stopped sharing data, and their employees lost access to one another’s internal systems.

That is reassuring as a separation measure, but it creates implementation work. Systems built during the acquisition period must be audited for credentials, dependencies, copied data, and shared development tools.

Engineers must determine which components belong to Manus and which rely on Meta services. Legal teams must establish what intellectual property can move with the separated company.

The timeline remains unclear. Operational separation can happen before contracts, code ownership, employee assignments, and financial obligations are fully resolved.

This uncertainty should temper claims that Manus has already regained independence. The company is moving toward independence, but the final business structure has not been announced.

For customers evaluating the Manus AI agent, ownership is only one factor. Reliability, security documentation, data location, and support continuity deserve equal weight.

Teams adopting any agent should keep exportable records of prompts, outputs, decisions, and source material. A searchable AI knowledge base can reduce dependence on one automation provider when a service changes ownership or access policies.

The Reversal Pressures Meta and Cross-Border AI Startups

Meta loses a shortcut into agentic AI, while globally structured startups lose confidence that relocation can insulate a transaction.

A Shopify catalog produced through the Manus integration demonstrates that the service continued shipping external workflows after operational separation began. It does not verify Manus’s final ownership structure. Source: Manus official blog, May 2026.

Meta acquired Manus while competing with OpenAI, Google, and Anthropic to turn AI models into products that perform useful work. The startup offered an established service, revenue, and engineering experience in autonomous task execution.

The unwind forces Meta to replace those benefits internally or through another acquisition. It cannot assume that the Manus team, technology, or customer activity will remain available.

Meta still has major advantages. It controls large consumer platforms, operates extensive computing infrastructure, and continues investing in models and AI talent.

However, distribution does not automatically produce a dependable agent. An agent must coordinate tools, maintain context, handle permissions, and recover when a task fails.

Manus had been working on those product problems before joining Meta. Losing that acquisition removes one route for accelerating Meta’s agent plans.

The outcome also affects Meta’s future deal screening. Any target with Chinese founders, investors, employees, technology, or earlier operating entities now presents a clearer unwind risk.

That does not make every such acquisition impossible. It raises the cost of confirming which regulators can intervene and what happens if they act after closing.

Startups face the opposite pressure. Founders often use international holding companies and relocate teams to access global customers and investors.

The Manus case shows that legal headquarters can be only one part of regulatory identity. Authorities can look through a Singapore parent to the company’s Chinese development history.

Investors must therefore examine technology-export rules and data-transfer exposure earlier. They also need contingency plans for returning capital, separating systems, and restoring ownership.

A forced unwind is harder for an AI company than for a business whose assets are mostly physical. Code, model improvements, employee knowledge, and training data can cross organizational boundaries during integration.

Once shared, some knowledge cannot be cleanly returned. The companies can revoke access and stop new transfers, but they cannot make employees forget technical information they already encountered.

That creates enforcement challenges. Regulators can require corporate separation, yet measuring whether all technological benefit has been reversed is much harder.

The original Meta acquisition also promised to end Chinese ownership interests in Manus and discontinue its services in China. The reported buyback points in the opposite direction.

That is the article’s central reversal. A transaction designed to remove Chinese ownership now appears likely to restore Chinese investors and potentially establish a China-incorporated joint venture.

The pressure extends to Tencent. Participating in the buyback could secure access to a commercially successful agent company, but it also requires considerable capital and regulatory coordination.

Tencent must avoid creating the impression that Manus is simply exchanging one dominant owner for another. Its strategic interest in agents makes that distinction difficult to maintain.

Manus carries the greatest operational pressure. It must convince regulators, investors, employees, and customers that it can survive a second major corporate transition within one year.

Revenue growth helps that argument. Missing information about margins, retention, and governance weakens it.

The company’s best defense is continued product performance. Stable service, transparent data policies, and reliable integrations would show that Manus can function outside Meta.

Service disruptions, unclear ownership, or changing regional availability would support the opposite interpretation. They would suggest that Manus’s independence exists more clearly in transaction documents than in daily operations.

Three Signals Will Show Whether Manus Is Truly Independent

The next phase will be decided by financing, governance, and customer continuity rather than another headline about negotiations.

The first signal is a signed transaction with a disclosed ownership structure. Reports that investors are considering a buyback establish intent, not completion.

A final announcement should identify the purchasing entities, Manus’s parent company, and the governance rights held by each major investor. It should also explain whether Tencent is a minority participant or the central source of control.

If the agreement preserves management authority and distributes investor influence, the independence case becomes stronger. An arrangement dominated by one strategic investor would weaken it.

The second signal is a detailed regulatory and corporate structure. Manus reportedly considered becoming part of a China-incorporated joint venture with a possible Hong Kong listing.

A confirmed structure would clarify how Beijing expects the company to handle technology, employees, and capital. It would also show whether Manus intends to operate separate regional services.

For international customers, the critical disclosures concern data storage, processing entities, cross-border transfers, and model access. A company can operate globally only if buyers understand which rules govern their information.

Clear documentation would reduce the uncertainty created by the unwind. Vague statements about independence without corresponding data policies would leave the central risk unresolved.

The third signal is sustained commercial performance after Meta’s support disappears. The reported revenue run rate is impressive, but it remains unaudited in public reporting.

Customer retention will matter more than short-term growth. Manus must show that subscribers continue using the product after the operational split and ownership changes.

Product releases are another useful indicator. Continued integrations and stable task performance would suggest that the engineering organization retained the resources it needs.

Watch for enterprise references, security certifications, and transparent service reporting. These signals provide stronger evidence than viral demonstrations or unsupported user counts.

The same test applies to the broader agent market. OpenAI, Google, Anthropic, Meta, and Tencent are all trying to connect AI models with real actions.

Manus’s unusual path shows that technical capability is only one part of that competition. Ownership, jurisdiction, data control, and distribution can determine whether an agent reaches users at scale.

The Manus AI agent is not independent yet in any final, documented sense. It has been operationally separated from Meta, and its former investors reportedly want to restore independent ownership.

That makes the company worth watching as both a product and a geopolitical test case. If Manus completes the buyback while retaining customers, it will show that an agent business can survive a forced separation from a major platform.

If financing stalls or governance shifts toward another dominant technology company, the independence narrative will look less convincing. Readers should watch the signed ownership terms, regional data policies, and post-separation customer performance before treating the reversal as complete.