The Deal, the Order, and the Reversal
Meta closed its acquisition of Manus, the Chinese-founded AI agent startup, on December 29, 2025, for a reported price of roughly $2 billion. Four months later, in April 2026, China's National Development and Reform Commission prohibited the acquisition and ordered the parties to unwind it, citing violations of rules on outbound investment and technology export controls. Chinese Foreign Ministry spokesperson Lin Jian said on April 28 that the government conducts reviews of foreign investment and makes decisions on it in accordance with law; a Chinese industry analyst, Ma Jihua, described the ruling as part of Beijing's continued effort to tighten its foreign-investment review system for cross-border mergers and acquisitions.
On August 11, 2026, Manus announced it would resume operating as an independent company, completing the separation process. In the intervening months, Meta had already begun cutting Manus off from its internal systems and barring its own staff from using Manus's tools, moving toward operational separation well before the public announcement. The deal, from signing to full reversal, spanned roughly eight months.
The Structure That Was Supposed to Work
Manus was incorporated offshore, reportedly in Singapore, a structure that acquirers of Chinese-founded technology companies have used for years on the theory that a foreign holding company sits outside Beijing's jurisdiction over domestic mergers and acquisitions. Critics have called this pattern Singapore washing: routing the legal ownership of a China-origin technology company through a third jurisdiction to sidestep Chinese regulatory review of the underlying business.
The NDRC's order made the limits of that theory explicit. Regulators treated the offshore incorporation as immaterial to the review, because the technology and the talent that created it originated in China regardless of where the parent entity was legally domiciled. For an acquirer, the lesson is specific rather than general: a holding structure changes where a company is registered, not where its underlying assets, code, models, and the people who built them, are treated as having originated for regulatory purposes.
The Personal Risk No Term Sheet Prices
Manus co-founders Xiao Hong, the chief executive, and Ji Yichao, the chief scientist, were summoned to Beijing in March 2026 as the review proceeded, and both were barred from leaving China during that period. That detail rarely appears in the standard risk factors an acquirer negotiates around in a cross-border technology deal, which typically focus on antitrust clearance, foreign-investment screening timelines, and integration cost. A regulatory review that can restrict named individuals' freedom of movement is a different category of risk entirely, one that lands on the founders personally rather than on the balance sheet of either company.
For any acquirer bringing on a founding team based inside the jurisdiction whose regulator might later object to the deal, that risk deserves its own line in diligence: what happens to the people, not just the company, if the review goes the wrong way, and does the deal structure or the founders' own mobility plans account for it.
The Checklist for the Next Cross-Border AI Deal
The Meta-Manus unwind is a compact case study in a risk class that is likely to recur as more AI acquisitions cross the US-China line, directly or through third-country structures. An offshore wrapper does not neutralize the exposure of a China-origin technology company to Chinese export-control and outbound-investment rules; the origin of the technology and the team is what NDRC-style review tests, not the location of the holding entity. Sequencing matters as much as structure: Meta closed in December and only learned the deal would be unwound in April, meaning a full quarter of integration, staffing changes, and data-system access decisions had to be reversed after the fact.
The reversal itself was not free. Cutting off system access, barring cross-use of tools, and deleting data created since the close are real operational costs that a clean-unwind clause on paper does not eliminate in practice. Any acquirer evaluating a similar structure should price in the cost of a full reversal, not just the cost of closing, and treat a regulator's willingness to act months after signing as the baseline case rather than the tail risk.
Read next: China Trained a Frontier AI With No US Chips | Anthropic Eyes $6B Deal to Cut Its Compute Bill
