Manus told its users on Tuesday that it will “soon return to operating as an independent company,” closing out one of the strangest sequences in recent AI dealmaking. Meta completed its roughly $2 billion acquisition of the agent startup on December 29, 2025. Chinese regulators ordered the deal unwound in April. Eight months after closing, the buyer is handing the company back.
The mechanism was China’s foreign investment security review, and reporting at the time described the reversal of a completed acquisition as unprecedented under it. The stated concerns, per CNBC and the South China Morning Post: Manus had not notified Chinese regulators before signing with Meta, and letting a homegrown agent company sell itself to a US platform risked setting a precedent for transferring Chinese AI technology to a strategic rival. Meta began dismantling the arrangement in June, cutting Manus off from its internal systems.
Manus was founded in China in 2022 and later relocated to Singapore — a structure that, on paper, was supposed to make exactly this kind of transaction clean. It did not.
The part that lands on users next week
There is an operational deadline attached. Data generated by certain users on or after December 29, 2025 — the day the Meta deal closed — will be deleted between August 23 and 24 Singapore time. Backups need to be finished before 7:59 a.m. SGT on August 23, with restoration opening August 25. Manus has been explicit that this is regulatory compliance, not a security incident, and that unaffected users can carry on normally.
That distinction matters legally and not at all practically. If you built a workflow on Manus during those eight months, the reason your data is disappearing is irrelevant to the fact that it is disappearing, on a fixed date, because of a corporate ownership dispute you were never party to.
Our take: The $2 billion is the least interesting number here. What just got established is that in China’s strategic sectors, a closed deal is not a finished deal — regulators demonstrated they will reach back through a completed transaction, force a divestiture, and impose data consequences on end users along the way. No amount of clever structuring priced that out; the Singapore domicile did not help. For anyone buying, integrating, or building on top of a Chinese-origin AI vendor, that is a new line item: reversibility risk, which behaves nothing like the regulatory risk most deal models already carry. It is also a warning about vendor concentration generally. The failure mode here was not the product breaking or the company running out of money. It was ownership becoming a geopolitical question. If a tool sits in a workflow you cannot afford to lose, the exit plan needs to survive causes that have nothing to do with the tool.
What to watch
- The August 23 deletion actually executing on schedule — and whether the scope of “certain users” widens between now and then.
- What Manus looks like standalone: who funds it, who holds the equity Meta gave up, and whether the product cadence survives the split.
- Whether Beijing applies the same review to a second inbound acquisition. One reversal is a case; two is a policy.
- Whether US acquirers now simply stop bidding on Chinese-origin AI assets, which would be the quiet, structural outcome.
- Meta’s accounting treatment of the unwind, and what it says about the agent strategy the acquisition was meant to accelerate.