China’s regulators have ordered Manus, the autonomous-agent startup acquired by Meta Platforms, to unwind that deal, Bloomberg reported on May 21. The three co-founders are now exploring a fundraise of approximately $1 billion from outside investors to finance a buyback, and may contribute their own capital to cover the remainder.

The founders, Xiao Hong, Ji Yichao, and Zhang Tao, are in talks about a new investment round at a valuation that would at least match the price Meta paid, according to people familiar with the discussions who asked not to be named on a private matter. If the fundraise closes, the plan is to reconstitute Manus as a Chinese joint venture with those new backers, then pursue a Hong Kong IPO.

A forced reversal of a completed acquisition at this scale is nearly without modern precedent. M&A deals do get unwound under regulatory pressure, but almost always before they close, under conditional clearance or blocking orders. Post-completion forced divestitures are reserved for the most acute antitrust or national-security violations, and even those unfold over years. The Manus situation appears to be moving faster, and the transaction involved no revenue dominance argument, only the geopolitical fact of a Chinese-founded AI company becoming a wholly owned subsidiary of a US platform.

The regulatory logic is not hard to reconstruct, even if Beijing has not publicly cited its reasoning. Manus is a Chinese-founded agentic AI company. Its product handles autonomous multi-step tasks, which means it processes sensitive instructions and potentially sensitive data on behalf of users. Chinese regulators have grown more attentive to outbound flows of AI capability and user data toward US parent companies. The Cyberspace Administration of China and the State Administration for Market Regulation have both expanded their cross-border data and M&A review frameworks since 2021. A deal structured as a full acquisition by Meta would give a US company control over the team, the training data, and the infrastructure. That combination fits the profile of what Beijing has signaled it will not permit.

This is not a one-directional constraint. In the United States, CFIUS (the Committee on Foreign Investment in the United States) has blocked and unwound inbound deals where a foreign acquirer gains access to sensitive technology or US person data. The Manus case is the mirror image: a Chinese regulator stopping an outbound deal where a US acquirer gains control of a Chinese-founded AI capability. Both governments are now applying national-security logic to AI acquisitions, which means the cross-border M&A market for AI companies has become structurally constrained from both ends at once.

Bloomberg’s reporting does not disclose the specific buyback valuation Manus must pay, whether Meta is contesting the order, or what timeline regulators have set. Those gaps matter. If Meta agreed to cooperate quickly, the deal could close without litigation. If Meta resists or demands a premium, the founders’ fundraising task becomes harder, because they would need to cover not just the original price but any negotiated exit premium on top of it.

The founders contributing their own money alongside outside investors is a structural signal worth noting. It aligns their incentives with new backers and signals confidence in the reconstituted company’s path to a Hong Kong listing. It also reflects a straightforward reality: raising $1 billion to buy back a company you previously sold is not a typical venture pitch. Investors are being asked to back a re-privatization into a regulatory-compliant structure, not a growth story from scratch.

Any founder or investor currently weighing a cross-border AI exit should treat this episode as an active data point. The assumption that a completed deal is a completed deal no longer holds when the acquired company has Chinese origins and the acquirer is a major US platform. Regulatory review on both sides has expanded well past the pre-signing stage, and the Manus situation demonstrates that “closed” is not always final.

Reported by Bloomberg on 2026-05-21.