Manus Founders' Travel Restrictions to Be Lifted; $20 Billion AI Deal Leaves Lasting Impact

For Manus, the most critical phase has passed, but that does not mean the risks have disappeared.

Author and source: 0x9999in1, ME News



• The travel restrictions on Xiao Hong and others are about to be lifted. What matters most is not the individuals' travel plans themselves, but the fact that the separation between Manus and Meta is nearing a state acceptable to regulators. What regulators truly require is the termination of Meta’s control and acquisition relationship with Manus—not that Manus abandon its internationalization or return to China.
• Manus continues to operate from Singapore, which is a crucial yet easily overlooked aspect of the entire situation. At least based on the current solution, “where the business operates” and “who ultimately controls the core assets” are treated as two separate issues.
• Tencent has become the largest single shareholder while maintaining a minority stake; Manus will not be absorbed into Tencent. This is not a traditional "big tech acquisition," but rather a capital structure shaped by practical constraints: Chinese capital has reemerged as a significant shareholder, while the company retains its independent overseas operations.
• For Manus, the most critical phase has passed, but that doesn’t mean the risks are gone. The final spin-off still requires regulatory approval, and some user data will undergo deletion, backup, and restoration procedures by late August, demonstrating that the legal and data ties from the Meta era must be genuinely severed—not merely renamed on corporate documents.
• The most significant industry implication of this transaction is that it establishes a much clearer boundary for all AI entrepreneurs with Chinese backgrounds: while a company’s registration location can be moved, markets can be globalized, and teams can be internationalized, the historical chains of technology, data, personnel, and assets will not automatically be erased simply because a Singapore business license has been obtained.
• More subtly, Manus has not lost its commercial value. When it was acquired by Meta at the end of last year, its annual recurring revenue had already exceeded $100 million; in June this year, media outlets citing insiders reported that its annualized revenue run rate once reached $400 million to $500 million. Regulation blocked a deal, but not the product itself.

If you only read the headline, this event could easily be interpreted as a simple statement: Xiao Hong can finally return to Singapore.

But if you rewind the clock by five months, the taste is completely different.

In March of this year, Xiao Hong, Ji Yichao, and other Manus management personnel were required to remain within the country. At that time, what confronted them was not an ordinary inquiry regarding a cross-border merger or acquisition, but rather a reassessment of an AI transaction that had already been completed, valued at over $2 billion, with Meta as the buyer. On January 8, the Ministry of Commerce publicly stated that it would collaborate with relevant departments to assess the transaction’s compliance with laws and regulations concerning export controls, technology import and export, and overseas investment. On April 27, the Office of the Foreign Investment Security Review Mechanism of the National Development and Reform Commission further issued a definitive conclusion: prohibiting the investment and requiring the parties to withdraw the acquisition transaction.

From "assessment" to "revocation," regulatory stance leaves little room for ambiguity.

So today, reports emerged that regulators are preparing to lift travel restrictions on Xiao Hong and others. The real signal to interpret is not that “no one is in charge”—on the contrary, it’s precisely because the core regulatory objectives are being achieved that conditions have now arisen to lift restrictions on individuals.

The logic between these two is very clear.

Meta has exited, and Manus has resumed independence; original shareholders have restructured capital, with Tencent becoming the largest single shareholder without gaining control; the company will continue to operate independently in Singapore; the related arrangement is pending final approval. As long as this arrangement is ultimately implemented, the most sensitive chain—the complete takeover of a cutting-edge AI company with Chinese technological and talent roots by a U.S. tech giant—will be severed.

In other words, Xiao Hong's departure does not mean the regulators have lost; a more accurate understanding is that the issues the regulators sought to address are being resolved in a direction acceptable to them.

This is why I believe the symbolic significance of lifting these outbound restrictions may be even greater than the restrictions themselves.

It clearly defined the boundaries of the Manus incident: regulators aimed to stop that specific transaction, not to prevent the team from continuing its global operations.

This is the most intriguing aspect of the entire matter.

Manus initially developed in China, relocated its operational focus to Singapore in 2025, and then joined Meta at the end of last December. Following the simplest business logic—if a company has moved, its core operations are overseas, and its customers are global—then it appears to already be an "overseas company."

The reality, however, is that things are not that simple.

The Ministry of Commerce's statement in January already made the issue very clear. Behind cross-border mergers and acquisitions are not just questions of “which company buys the equity of which company,” but also involve whether technology has been transferred overseas, whether data has crossed borders, whether outbound investment procedures are compliant, and whether the entire transaction chain adheres to relevant regulations.

This means that for strategic technology industries like AI, the place of registration is not a universal eraser.

You can relocate the company’s legal entity to Singapore, but you cannot erase, merely by moving the location, the origins of the technology developed by the team, the history of the R&D personnel, the process of data and intellectual property creation, or the pathways of asset transfers between domestic and international jurisdictions.

This is likely the most important institutional legacy of the Manus incident.

However, looking at it the other way, regulators ultimately did not require Manus to return to China to operate. At least according to the current arrangement disclosed by the Financial Times, the company will remain based in Singapore. Although Tencent is expected to become the largest single shareholder, it will hold only a minority stake and will not consolidate Manus as a Tencent business unit.

This detail is important because it shows that the situation hasn't escalated to the extreme of "since it has Chinese ties, it must return to China."

Control over capital is one thing; global operations are another.

For an AI company, this distinction is immensely valuable. Manus has users worldwide, and Singapore serves not only as its headquarters but also as the hub for international fundraising, talent acquisition, customer expansion, and data infrastructure arrangements. If regulators ultimately accept this structure, the message at least in Manus’s case is clear: globalization itself is not rejected—what is being redefined is control and compliance within the globalization process.

This issue is more complex than simply asking whether one can go overseas, yet it is far more important than a flat “no, you cannot go overseas.”

Next, we can't avoid Tencent.

According to the currently disclosed plan, existing stakeholders including Tencent, ZhenFund, HSG, and management will participate in the buyback, while Benchmark will not continue its involvement; Tencent will assume the majority of Benchmark’s shares. Upon completion of the transaction, Tencent is expected to become the largest single shareholder of Manus but will remain a minority shareholder, not exceeding the control threshold.

Seeing this, it’s easy to wonder: Has Tencent finally acquired Manus?

No.

Currently, there is no reliable information supporting the claim that Manus will be controlled by Tencent or that Manus technology will be fully integrated into Tencent. On the contrary, the existing proposals repeatedly emphasize independent operations.

But that doesn't mean what Tencent received has no value.

In my view, what it received is more like an expensive AI option.

This year, Tencent is increasing its investment in AI more than ever before. Tencent's latest second-quarter revenue reached RMB 204.8 billion, a 11% year-over-year growth; capital expenditures for the quarter have already reached RMB 52.8 billion, a 176% increase year-over-year, significantly higher than the same period last year. The company's management has clearly stated that the large-scale procurement of computing power is aimed at converting future model and application usage into revenue.

The question is, what is truly scarce in the age of AI?

Is it only GPU? Obviously not.

As large models become increasingly accessible and their core capabilities increasingly commoditized, what’s truly hard to acquire is shifting to product definition, user habits, workflow entry points, and sustained monetization. Manus has already demonstrated something only a handful of agent startups have proven: users are indeed willing to pay directly for a general-purpose agent.

In December last year, Manus announced that its annual recurring revenue exceeded $100 million. By June this year, media outlets citing informed sources reported that its annualized revenue run rate had grown to between $400 million and $500 million. Of course, this is not full-year revenue in an audit sense, nor is it profit; ARR during a period of rapid growth cannot be simply equated to final realized cash flow.

Even with a significant discount, it still shows that Manus has crossed the hardest hurdle for AI startups: moving from "everyone thinks it's cool" to "people are consistently paying."

This is why a buyback price of around $2 billion doesn't seem as crazy as it might appear.

At a $100 million ARR, Meta’s initial valuation was approximately 20 times revenue; if we use the reported run rate of $400 million to $500 million as of June this year, the same $2 billion valuation would quickly drop to a multiple of around 4 to 5 times. Of course, no one can guarantee whether the company will maintain such growth rates after the split—the latest report from the Financial Times offers a more cautious outlook, estimating that the standalone annual recurring revenue will still exceed $300 million.

What truly matters isn’t precisely calculating a multiple, but recognizing the shift: Six months later, Manus is no longer just a story backed by concepts with a $2 billion valuation. If these revenue figures ultimately prove sustainable, the company’s fundamentals are now catching up to what once seemed an extremely aggressive price.

Tencent becoming the largest shareholder essentially means keeping a strong hand—already featuring a product, users, revenue, and global potential—on the table in a sector where it has already made significant investments.

But it didn't take the cards directly back into its own hand.

This might just be the smartest thing right now.

What about Meta?

At first glance, spending over $2 billion to acquire a company only to be forced to dismantle it six months later would not be a pleasant merger and acquisition story.

Especially for Meta, what it originally purchased was not just a group of engineers.

In December last year, Manus officially disclosed that since its launch, it has processed over 147 trillion tokens and enabled the creation of over 80 million virtual computing environments. What Meta truly values is an already operational Agent product, along with the proven product pipeline from model to execution.

For an internet giant betting on "superintelligence" and AI agents, the greatest value of this capability is actually time.

Can you do it yourself? Of course you can.

The question is, someone else has already been running for a year—how much would you be willing to pay to skip half a year?

Meta's previous answer was over $2 billion.

But from another perspective, this loss shouldn’t be portrayed as a major setback to Meta’s AI strategy. In the second quarter of this year, Meta already narrowed its 2026 capital expenditure guidance to $130 billion to $145 billion, with AI infrastructure and investments in the Superintelligence Lab remaining the primary drivers. Compared to this figure, the $2 billion Manus transaction is strategically significant but not financially decisive.

Meta has sufficient capital, computing power, talent, and a global distribution network to rebuild its agent capabilities.

What cannot be easily made up for with money is time and organizational experience.

Manus has already encountered various failure scenarios arising from real users employing Agents for research, coding, file handling, and executing complex tasks. These issues won’t be fully documented in papers, nor will they automatically disappear by purchasing tens of thousands more GPUs.

So the real loss for Meta from this deal falling through isn't "losing a company," but rather losing a cutting-edge product team that had already navigated numerous challenges in the real business world on its behalf.

Today, in the AI industry, the most expensive things are often precisely these pitfalls.

However, there's no need to rush into writing a "resounding victory" ending for Manus.

It indeed saved the company, the founder's travel restrictions are set to be lifted, the Singapore operational base is preserved, and once the capital structure is stabilized again, Tencent could become a highly influential shareholder.

It sounds like a narrow escape.

However, the challenges Manus faces next may be even greater than before being acquired by Meta.

Because from now on, it can no longer rely on Meta’s balance sheet to speak to the future.

Even though the Meta era was brief, it meant that Manus once had access to one of the world’s most powerful computing resources, distribution channels, advertising systems, and AI talent platforms as potential backing. Now, as an independent company again, every dollar of inference cost, every round of funding, every customer acquisition, and every enterprise client contract has returned to the books of a startup.

At this point, $400 million or $500 million in annual recurring revenue is no longer the end of the story, but the beginning of pressure.

High-growth Agent companies eventually hit a wall: completing tasks requires extensive model calls and computational resources, while the price users are willing to pay doesn’t grow indefinitely. If revenue increases rapidly but token costs rise even faster, the business model remains at risk.

So what truly matters in Manus’s next phase isn’t how much further its valuation can rise, but three more fundamental questions: How is user retention performing? Can inference costs continue to decline? And can enterprise customers form a stable, high-margin revenue structure?

How much is an agent that can deliver a stunning demo worth?

The answer could be tens of billions of dollars.

How much is an Agent worth that can reliably execute tasks in front of millions of users and generate profit with every execution?

That might be an entirely different company.

Manus has not yet fully answered the second question.

There’s another detail, easily overlooked beneath the headline “Xiao Hong is about to return to Singapore,” that is especially worth noting.

Manus has notified certain users that, due to the company's separation from Meta and resumption of independent operations, data generated for some users after Meta's acquisition on December 29, 2025, must be deleted between August 23 and 24 this year. Affected users can back up their data in advance and restore it starting August 25.

Why would a company that has regained its independence go through such a complicated data migration process?

Because a true split is never just about updating the shareholder register.

Capital can be transferred back with a single agreement, but data cannot.

What data was created after Meta took control? Where is it stored? Who had access to it previously? Can the new independent entity legally retain it? What must be deleted and can only be restored by user action?

These are the most troublesome underground pipelines for tech companies to split.

Precisely for this reason, Manus’s “reindependence” is not just public relations language. At least based on public actions, both parties are actively separating their data and operations.

In fact, the data deletion on August 23 speaks louder than any statement.

Meta must truly leave.

Not a departure in terms of branding, but a separation in terms of data relationships, operational relationships, and control relationships.

Will this upheaval ultimately change the trajectory of China’s AI startups?

Absolutely.

Over the past two years, many entrepreneurs held a simple but not unreasonable belief: if domestic competition is too fierce, relocate the company’s headquarters to Singapore; if raising funds in U.S. dollars becomes increasingly difficult, internationalize the company in advance; and if the primary customers are all overseas, build a global market from day one.

Manus was once the most brilliant example on this path.

In 2025, the startup world was still debating whether “agents” had a viable business model—by year’s end, Meta acquired the company for over $2 billion. For founders and early investors, this was nearly a textbook example of a global exit.

Then regulators told the entire industry: the map isn't this simple.

In the future, when an AI company with Chinese R&D and personnel roots undergoes a cross-border restructuring, investors will no longer ask only, “Cayman or Singapore?” or “USD fund or RMB fund?”

They must continue asking.

When were the key technologies developed? Which entity are the core employees employed by? How is intellectual property transferred? Where do the training data come from? Have the data been transferred overseas? What authorizations have been made between domestic and overseas entities? Who is the final buyer? Who holds actual control after the transaction is completed?

Once these issues are introduced into financing and M&A models, transaction costs across the entire industry will certainly rise.

But this does not mean that Chinese AI companies' globalization will stop.

Manus ultimately remained in Singapore, which itself serves as counter-evidence.

What is likely to truly come to an end is not overseas expansion itself, but the crude approach of “move out first, deal with historical relationships later.”

Future globalization will be more expensive, slower, and more lawyer-driven, requiring technology, data, intellectual property, and personnel structures to be designed from day one of company formation.

This is certainly not good news for entrepreneurs.

For an industry that has already entered the core of great power technological competition, it is difficult to expect AI cross-border transactions to remain as simple as they were during the consumer internet era.

So, what do you think about today’s news?

I don't believe this represents a 180-degree shift in regulatory stance, nor do I think it means Manus has fully cleared all hurdles.

A more accurate description is: Both parties are arriving at a mutually acceptable position, though neither has fully obtained what they originally wanted.

Meta lost Manus.

Regulators have dismantled the control of U.S. tech giants over this China-linked AI company.

Tencent has re-entered the list of core shareholders without directly acquiring Manus.

Manus preserved itself, as well as the Singapore and global markets.

Xiao Hong paid with an extremely costly period of time and uncertainty.

This is not a complete victory for either side.

Sometimes in the business world, the most valuable outcome isn't a complete victory, but rather ensuring a company can continue moving forward.

A year ago, people discussed Manus because it was seen as potentially becoming the first truly commercially scaled general agent; six months ago, discussions centered on why a Chinese team could sell it to Meta for over $2 billion; in the past few months, the conversation has shifted to whether this company can even recover.

The answer is becoming clearer now.

Manus is back.

But it's no longer the Manus from last year.

It no longer has Meta behind it, the shareholder list has been reshuffled, the regulatory boundaries have been redrawn, data is being segmented, and the company has once again become a startup that must prove its business model on its own.

If Xiao Hong returns to Singapore safely, what may truly be cause for celebration is not finally boarding that plane.

Instead, after the plane lands, this company still has another round to play.

Whether this round can be won no longer depends on a $2 billion transaction or a regulatory showdown.

In the end, it still comes down to the most brutal—and fairest—question in the tech industry:

Is the product truly good enough, and are users willing to keep paying for it?

1. Financial Times, "China Poised to Lift Travel Ban on Manus Founders," August 15, 2026. Regarding the lifting of travel restrictions on Xiao Hong and Ji Yichu, Tencent and original shareholders' repurchase arrangements, Manus's continued independent operation in Singapore, and the final approval status.
2. Manus, A Note to Our Users, August 11, 2026. Manus confirms it will resume independent operations and discloses the schedule for partial user data deletion, backup, and restoration from August 23 to 25.
3. Office of the Working Mechanism for Security Review of Foreign Investment, National Development and Reform Commission, "Decision on the Security Review of Foreign Acquisition of the Manus Project," April 27, 2026.
4. Ministry of Commerce of the People's Republic of China, "Ministry of Commerce Holds Regular Press Conference," January 8, 2026. Regarding compliance assessments related to export controls, technology import and export, and foreign investment concerning Meta's acquisition of Manus.
5. Reuters, "Manus Original Investors Plan to Buy Back AI Firm from Meta for $2 Billion," June 18, 2026. Report on the $2 billion buyback plan and Manus's annualized revenue run rate increasing from approximately $100 million to $400–500 million.
6. Reuters, "Tencent in talks to become AI startup Manus' largest shareholder," July 10, 2026. Report on Tencent, ZhenFund, and HSG participating in the buyback, and Tencent's plan to become the largest shareholder.
7. Meta Platforms, "Meta Reports Second Quarter 2026 Results," July 29, 2026. Meta has set its 2026 capital expenditure guidance at $130 billion to $145 billion.
8. Reuters, "Tencent Q2 Revenue Rises 11% on AI-Driven Ad Gains, But Profit Falls Short," August 12, 2026. Tencent's Q2 revenue of RMB 204.8 billion, capital expenditures of RMB 52.8 billion, and AI investment details.