Some technology stories begin with a product launch. Others begin with a funding round, a benchmark, or a founder posting something dramatic at two in the morning.
This one begins with two executives being unable to leave a country while their company tries to reverse a $2 billion acquisition.
According to the Financial Times, Beijing is preparing to lift travel restrictions imposed on Manus co-founder and chief executive Xiao Hong and members of the company's management, including chief scientist Ji Yichao. Xiao reportedly plans to return to Singapore, where Manus is based, once the separation from Meta is resolved.
That sounds like the final chapter of an unusually complicated acquisition. I think it is more useful to see it as the first clear example of a problem many AI startups will face:
A company can move its headquarters, contracts, and employees. Its technology may still carry the regulatory history of where it was created.
Manus moved its headquarters and core engineers from China to Singapore before Meta acquired it. The transaction was still examined through its Chinese roots, investors, entities, talent, technology, and earlier operations.
You can move the company address. Apparently, you cannot simply git mv its jurisdiction.
How a $2 billion acquisition reached reverse gear
Manus is best known for its general-purpose AI agent: software designed to plan and complete multi-step tasks rather than only answer a prompt. Meta acquired the company in December 2025, only months after Manus moved its headquarters and key engineers to Singapore.
At the time, the fit looked logical. Manus had a fast-growing agent product, while Meta had billions of users, business customers, advertising systems, and a very large appetite for AI. In its original announcement, Manus said it would continue operating its subscription service and remain based in Singapore.
Then the regulatory problem grew.
China's Ministry of Commerce said in January that it would assess whether the acquisition complied with rules covering export controls, technology transfers, overseas investment, data movement, and cross-border M&A. The ministry's position was quite broad: companies are free to operate internationally, but they still have to follow the relevant Chinese procedures. That statement is available through China's State Council Information Office.
In March, Xiao and Ji were summoned to Beijing and restricted from travelling abroad while the deal was reviewed. The FT reported at the time that no formal investigation had been opened and no charges had been brought.
In April, the National Development and Reform Commission blocked the takeover and ordered the transaction to be unwound. A useful legal analysis from DaHui Lawyers describes it as the first transaction-specific prohibition made public under China's foreign investment security review system.
By August, Manus was preparing to operate independently again. Reuters reported that the separation would also involve deleting some user data.
The simplified timeline looks like this:
| Date | What happened |
|---|---|
| December 2025 | Meta acquired Manus, reportedly at a valuation of about $2 billion. |
| January 2026 | China's commerce ministry opened a compliance assessment. |
| March 2026 | Xiao Hong and Ji Yichao were questioned in Beijing and prevented from leaving China. |
| April 2026 | The NDRC blocked the acquisition and required the parties to unwind it. |
| August 2026 | Manus announced a return to independence as investors worked on a buyback. |
| Now | Beijing is reportedly preparing to lift the travel restrictions once the unwind satisfies regulators. |
That is a remarkable amount of corporate history for a product that only became widely known in 2025.
Singapore did not erase the China connection
The most important detail is not that Manus operates from Singapore. It is that this fact was not enough to remove the transaction from Chinese scrutiny.
For years, Singapore has been a natural base for technology companies connecting China, Southeast Asia, the United States, and global investors. It offers access to capital, strong infrastructure, a respected legal system, and an international talent market. Moving there is not suspicious by itself; for many companies, it is simply a sensible business decision.
But regulators do not look only at the address printed below the logo.
They can look through the corporate structure and ask harder questions:
Where was the core technology developed?
Which legal entities employed the original team?
Where did the intellectual property come from?
Did technology, data, or technical personnel cross a border?
Which investors and founders still have economic control?
Was regulatory approval required before the transaction closed?
For a normal software company, these questions can already become complicated. For an AI company, the list grows quickly because the valuable asset is not just a code repository. It may include model behavior, agent architecture, training methods, evaluation data, workflows, research knowledge, and a small group of people who know how everything actually works.
That is why this case matters beyond Manus. The product was legally based in Singapore, but its origins remained relevant.
An offshore holding company is a structure. It is not a time machine.
In AI, the people are part of the technology
The travel restrictions make this story unusually human.
In most acquisition disputes, we talk about shares, approvals, intellectual property, and closing conditions. Here, the founders' ability to return to their company's headquarters became connected to the resolution of the deal itself.
I am not going to pretend we know every private conversation between Manus, Meta, and the Chinese authorities. We do not. The latest reporting relies partly on people familiar with the matter, and the final unwind still requires regulatory approval.
But the case highlights something easy to miss when discussing AI as if it were only compute and datasets: the team is a strategic asset.
Agent products depend heavily on product judgment, orchestration techniques, infrastructure choices, safety systems, and thousands of small decisions that may never be fully documented. Buying the company without retaining the people can be like buying a complicated machine without the engineers who know why it makes that noise every Thursday.
Governments understand this. So do acquirers.
When countries treat advanced AI as strategic technology, restrictions may affect not only chips, models, and source code, but also investment, data flows, ownership, and the movement of key personnel.
The company survived the deal
The surprising part is how active Manus remained throughout the dispute.
The FT reports that the platform continued receiving updates during the review and is expected to launch Manus 2.0 after the separation is completed. Its annual recurring revenue is expected to remain above $300 million after leaving Meta.
If those figures hold, Manus is not returning to independence as a broken asset waiting to be rescued. It is returning as a substantial AI business with a functioning product, customers, revenue, and a roadmap.
The proposed buyback also says a lot.
According to the FT, former investors including Tencent, ZhenFund, and HSG, together with management, plan to buy Manus back from Meta at approximately the same $2 billion valuation. Benchmark and some smaller investors are not expected to participate. Tencent would acquire much of that position and become the largest shareholder, but with only a minority stake. Manus would continue operating independently from Singapore rather than being absorbed into Tencent.
This is not a simple return to the old cap table. It is an attempt to create a new ownership structure that regulators can accept without destroying the operating company.
That distinction matters. Regulators are not necessarily saying that Manus cannot become a global business. The message appears closer to: it can operate globally, but not through a transaction that authorities believe bypassed required rules.
Harsh? Yes. Random? Not really.
The lesson for founders: compliance is product architecture
Founders often treat regulatory work as something that arrives later, somewhere between a large funding round and the first enterprise customer asking for a 97-page security questionnaire.
For a global AI startup, that approach is becoming dangerous.
Compliance decisions can affect the company architecture from day one:
Know where every important asset came from
It is not enough to know which entity currently owns the IP. A buyer may need a defensible history of where the technology was built, who contributed to it, how rights were transferred, and which export or investment rules may apply.
Map the team, not just the cap table
A company may be incorporated in one country while critical researchers, engineers, contractors, and previous entities remain connected to another. Those connections can determine which regulators consider themselves relevant.
Treat data separation as a real engineering requirement
The Manus–Meta unwind reportedly includes stopping data sharing and deleting some records. That is much easier when systems have clear ownership boundaries, access controls, retention policies, and export paths.
If two businesses are integrated by copying everything into one giant bucket called final_final_data_v3, separation will be memorable for all the wrong reasons.
Design the failure path before signing
Cross-border acquisitions need more than a happy path. Agreements and technical plans should answer what happens if approval is delayed, prohibited, or reversed after integration begins.
Who continues serving customers? Which systems must be disconnected? Who owns improvements made during the transition? What happens to employee access, customer records, subscriptions, and model integrations?
Those questions sound pessimistic until a $2 billion transaction has to travel backwards.
The lesson for buyers: due diligence is now geopolitical
For buyers, especially US technology companies, standard legal due diligence is no longer enough when acquiring an AI business with international roots.
The technical and regulatory review must happen together.
A company can have clean contracts, a Singapore headquarters, and an attractive product while still carrying approval risk from another jurisdiction. The more strategically important the technology, the less useful it becomes to ask only where the seller is incorporated today.
Buyers need to model at least three layers:
Corporate jurisdiction — where the entities are registered and where the deal is signed.
Operational jurisdiction — where employees, servers, customers, and data are located.
Historical jurisdiction — where the technology, team, funding relationships, and intellectual property originated.
The third layer is the one most likely to be underestimated.
This does not make international AI acquisitions impossible. It makes early regulatory analysis as important as valuation, security review, and product fit. The Manus case shows the cost of discovering that after integration has already started.
This is bigger than one acquisition
The case arrives during a wider contest over AI capability between China and the United States. Both countries increasingly treat advanced technology as an economic and national-security asset. The US uses chip restrictions, investment controls, and entity lists. China uses its own investment, technology export, data, and security-review systems.
It would be easy to reduce the Manus story to “China blocked an American buyer.” That is true, but incomplete.
The more lasting signal is that AI companies with Chinese roots may remain within Beijing's regulatory view even after moving their headquarters abroad. A legal relocation can change many things; it may not remove oversight of technology or talent developed earlier.
Other founders will notice. So will venture funds and potential acquirers.
Some startups may seek approval earlier. Some buyers may demand stronger closing conditions. Some investors may discount companies with complicated cross-border histories. Others may separate teams, data, and IP more deliberately long before a deal appears.
The result could be a more fragmented AI market in which companies choose their geopolitical lane earlier than they would prefer.
That is bad for the dream of technology moving freely across borders. It is also increasingly close to reality.
What happens next
As of August 16, 2026, several important pieces are still unfinished.
The buyback and separation need final regulatory approval. Xiao Hong's reported return to Singapore depends on the resolution proceeding as expected. Meta and Manus still have to complete the operational separation, including whatever data deletion and system changes are required. Manus then has to prove that its revenue, customers, and product momentum survive outside Meta.
If the unwind is completed, the result will be unusual: Meta will have bought an AI company, begun integrating it, and then sold it back at roughly the same valuation—all within months.
Manus, meanwhile, may emerge independent, well funded, and still growing. That would be a strong outcome after a process that could easily have damaged the company far more.
But the cleanest lesson belongs to everyone else.
A global company is not defined by a single headquarters pin on a map. It is a combination of code, people, data, ownership, history, and jurisdictions. In AI, every one of those components can become strategic.
The Manus deal did not collapse because the product stopped working.
It collapsed because the map around the product mattered more than the acquisition agreement expected.
Sources and further reading
Financial Times: China poised to lift travel ban on Manus founders
Reuters: Manus to resume independent operations as Meta deal unwinds
China's State Council Information Office: Commerce ministry comments on the investigation
DaHui Lawyers: China's first public foreign investment security review prohibition
If it annoys you twice, turn it into a tool.
See you in the next build.
— Ballwictb
Originally published on ZyVOP
💡 For more articles like this, subscribe to the ZyVOP newsletter!
Top comments (0)