In 1972, the meteorologist Edward Lorenz delivered a paper at the American Association for the Advancement of Science under a title that has long outlived the talk: “Predictability: Does the Flap of a Butterfly’s Wings in Brazil Set Off a Tornado in Texas?” Lorenz had earlier used a seagull. The butterfly stuck. Small perturbations, in a sufficiently complex system, produce disproportionate consequences. It was a useful image for atmospheric science. It was an irresistible one for venture capital.
Manus (the artificial intelligence start-up that became China’s most-discussed AI product of 2025 after DeepSeek) was a subsidiary of a parent firm named, without apparent irony, Butterfly Effect.
The basic facts are settled. Butterfly Effect was founded in Beijing in 2022. In May 2025, Benchmark Capital led a $75 million Series B at a $500 million valuation. By July, Manus had shut its Beijing office, retained only core technical personnel, and re-incorporated its parent in Singapore.
In December, Meta announced an acquisition of around $2 billion (the Wall Street Journal puts the figure at $2.5 billion), Meta’s third-largest deal ever.
Days later, China’s Ministry of Commerce opened a probe. In March 2026, the two co-founders, Xiao Hong and Ji Yichao, were summoned to Beijing and barred from leaving the country. On 27 April, the NDRC’s Office of the Working Mechanism for Security Review of Foreign Investment (invoking, for the first time publicly, a 15-year-old statute) ordered the deal unwound. By then the investors had been paid out. Benchmark had distributed proceeds to its limited partners. The Manus engineers had moved into Meta’s Singapore offices. The code had been integrated.
That a $2-billion transaction can be reversed at all is a striking proposition. That it has been ordered reversed ex post, months after closing, is more striking still.
The implications
What does it mean? Several things, in ascending order of consequence.
First, the geography of jurisdiction has been redrawn. The place of incorporation of the target no longer fixes the limit of regulatory analysis. The origin of the technology, the location of core R&D, the nationality of the founding team, historical operations in China, data flows, and the offshore restructuring process, all are now within reach. The Global Times editorial of April 28 defends the decision as a conventional security review and rejects the label of “long-arm jurisdiction.” (Beijing has good reason to dislike that phrase, having long deplored its American counterpart.) The label is contested, but the practice is not.
Second, the so-called Singapore-washing template is dead, or, more precisely, has been served notice. For two years, a steady migration of Chinese AI, fintech and crypto firms to the city-state has been tracked by anyone with a billable Mandarin speaker. The premise was simple. Re-register elsewhere, raise foreign capital, recruit globally, escape the Chinese regulatory orbit. The premise has not survived contact with the NDRC. Beijing has signalled, through Manus, that re-incorporation does not amount to escape.
Third, the symmetry with Washington is now almost complete. The US has spent five years restricting American investment in Chinese AI. Barring frontier semiconductor exports. Entity List has been expanded. And the Committee on Foreign Investment in the US has blocked or modified deals from TikTok to Magnachip.
Beijing has now demonstrated the mirror, using its own foreign investment security review, dormant for 15 years, to unwind a deal in which the acquirer is American and the target Singaporean. When two great powers reach for the same instruments, the instruments themselves become the language of the relationship.
Fourth, artificial intelligence has been quietly re-classified in China as a national security asset rather than an industrial one. (We won’t get into the parallel restrictions placed on Moonshot AI, Stepfun and ByteDance in this column, but save those for later.) Once a sector crosses that threshold, ordinary M&A logic (competitive pricing, exit liquidity, founder optionality) gives way to a different calculus. The Foreign Trade Law, the technologies catalogue, the foreign investment security review rules, these were always available. They are now available and willing.
Setting a precedent
Fifth, and most consequentially, the precedent will travel. CK Hutchison’s $23 billion port sale to a BlackRock-led consortium has been on hold since Beijing objected last year, and now waits for a Chinese partner to be inserted. Manus is the second high-profile cross-border transaction Beijing has refused to let lie. There will be a third. There will be a fourth.
There is a deeper question that sits behind all of these. Humpty Dumpty, in Through the Looking-Glass, told Alice that the only real question was which was to be master. Who is master of a piece of code, of a team of engineers, of a corporate domicile, once those things have crossed borders, that is not a question the international system has yet answered. It will answer it the only way it can. By accumulation of cases. Manus is one such case. A summit between the two presidents follows in mid-May. The agenda has been set, even if no one will say so.
A small footnote. The parent company was, as it turns out, named accurately. A flap of regulatory wings in Beijing has set off something resembling a tornado in Menlo Park.
Sinha writes on macroeconomics and geopolitics
Published on April 30, 2026