SINGAPORE, 15 AUG 2026 — Manus is becoming an independent company again. Meta is unwinding the US$2 billion acquisition it completed on 29 December 2025, and users who joined on or after that date have until 23 August to download their data before it is deleted.
The deal did not fail commercially; Chinese regulators blocked it. What should concern founders in this region is where Manus was incorporated when that happened, which was Singapore.
How it unwound
- Butterfly Effect founded
In China, with offices in Beijing and Wuhan.
- Manus launches
The agent product goes viral on a demo video. Annualised recurring revenue later passes US$100 million.
- US$75m round led by Benchmark
American venture money arrives. The company then closes its China offices, cuts staff, removes its Chinese online presence and reincorporates in Singapore.
- Meta acquires Manus
US$2 billion, to strengthen Meta's work on AI agents.
- Beijing blocks the deal
Reported as 27 April, under China's foreign investment security review mechanism.
- Users notified
Manus tells users it will soon operate independently again, and that data generated on or after 29 December 2025 by certain users will be deleted.
- Backup deadline
Deletion follows, with accounts reported to be restored from 25 August.
What the Singapore move was for
The 2025 sequence is not subtle. Take American money, then leave China on paper.
United States investment restrictions on Chinese AI companies made the original structure a problem for a fund like Benchmark. Reincorporating in Singapore solved that problem. It gave the company a neutral flag, a jurisdiction American investors are comfortable with, and a corporate identity that did not read as Chinese to a Washington regulator.
It worked for its stated purpose. The round closed, and a little over half a year later Meta paid US$2 billion for the result.
The move did not, however, change where the technology and the people came from. Chinese regulators took the view that both originated in China and remained within Beijing's authority regardless of the certificate of incorporation, and that view is what stopped the sale.
The doctrine underneath
The rule Beijing is asserting is that control follows origin rather than registration.
This is a broader claim than export controls make. Export controls attach to goods and to the nationality of the parties. The rule being applied here attaches to a company's history, meaning its founding, its engineering and its engineers. That attachment survives a change of domicile, staff or website.
That puts Beijing's position in substantive agreement with Washington's. American restrictions already look past the paperwork to ask whether a company is meaningfully Chinese. Beijing has done the same thing in reverse. A company that redomiciles to escape one is not thereby outside the other, and it can end up inside both.
Manus was caught. It was Singaporean enough to take American capital, but still Chinese enough for Beijing to veto the sale.
What this means for companies here
Singapore has spent fifteen years being the place a regional company moves to, on reasons that are good and mostly unchanged. English common law and enforceable contracts sit under everything else. The professional-services bench is deep, the tax treaty network is wide, and international investors understand the jurisdiction before anyone explains it to them.
None of that is diminished by this case. What does change is the viability of one specific tactic, which is redomiciling to scrub a company's origins ahead of somebody else's national-security review.
The distinction matters because the two uses have been quietly merged in a lot of founder conversations. Moving to Singapore for its legal system and capital markets still works. Using the move to obscure a company's origins for a future acquirer, however, has now failed a public US$2 billion test.
For Singapore's regulators this is a mixed outcome. The country benefits from being a neutral venue, not from being read as a laundry for corporate nationality. That perception invites the kind of look-through scrutiny that makes neutrality less useful for everyone incorporated here.
The practical advice for a founder is narrow. If your engineering history sits in a jurisdiction that asserts authority over its technology, a holding company elsewhere is a tax and governance decision, not a shield. Price the exit accordingly, and ask early which capitals have to say yes.
The part users actually feel
Underneath the corporate structure, a functioning product has customers, and they are the ones being handed a deadline.
Anyone whose data was generated on or after 29 December 2025 has until 23 August to back it up. Manus attributed the deletion to compliance with regulatory requirements in particular jurisdictions rather than to a technical constraint, which reads as the unwind requiring that data created under Meta's ownership does not travel to the independent company.
The position is legally coherent, but it makes for a poor experience. Eight months of an agent's working output — the artefacts, the intermediate steps, the accumulated context that makes an agent product worth paying for — is being removed on a twelve-day notice because of an ownership dispute the user has no part in.
It is also a reminder about agent products specifically. A chatbot transcript is a record of a conversation. An agent's history is closer to a working file, and the value of the tool grows with it. Anyone buying an agent product needs to ask where that working file lives and whose permission is needed to move it. Otherwise the context you have built becomes the lock-in.
Who ends up owning it
Reports on who ends up owning Manus are not consistent.
One line of reporting has the founders — Xiao Hong, Ji Yichao and Zhang Tao — exploring roughly US$1 billion from outside investors to buy the company back at the valuation Meta paid, with a Hong Kong listing as a longer-term outcome. Another has a Chinese consortium including Tencent, Sequoia China and ZhenFund funding the buyback, with Tencent taking the largest single stake while remaining a minority holder. At least one account describes that consortium as having completed the purchase and reclaimed control, which is a materially stronger claim than the others make.
These are not all compatible, none is confirmed by the companies, and the difference between them matters — a founder-controlled buyback and a Tencent-anchored consortium are different futures for the product.
What we could not establish
Which agency issued the block, on what statutory basis, and whether a formal decision was published. The reported date of 27 April 2026 and the attribution to a foreign investment security review come from secondary accounts rather than from a document we could read.
Also unestablished: what Meta recovers and on what terms, whether Manus keeps the model weights and infrastructure built under Meta's ownership, how many users the deletion affects, and whether the company will operate from Singapore or return to China. Neither Meta nor Manus has commented beyond the user notice.
What to watch
The ownership outcome, because it decides which market this product serves. A founder buyback with a Hong Kong listing points one way; a Tencent-anchored consortium points another.
Whether any Chinese-origin, offshore-incorporated company completes a foreign sale after this. One block tells you something about Manus, and it would take a second before anyone could call the doctrine settled policy.
And whether Singapore's own agencies say anything. A company incorporated here had its sale vetoed by another state on the basis that the incorporation did not count. That is a direct question about the value of a Singapore domicile, and so far it is not being asked out loud.
