China's National Development and Reform Commission ordered Meta to unwind its roughly $2 billion acquisition of agentic AI startup Manus in April 2026, blocking a deal agreed in December 2025 and signalling that Beijing is now prepared to use regulatory authority not just to protect domestic markets from foreign competition but to prevent strategic AI technology from flowing to American companies even when the target has already relocated to Singapore.
The Manus story starts as a startup success and ends as something more complicated. Manus was founded in 2022 by Hong Ji and Tao Zhang, with its parent company Butterfly Effect established in Beijing before the founders relocated to Singapore around mid-2025. The platform attracted global attention when it launched as one of the first general-purpose autonomous AI agents capable of completing complex tasks including market research, coding, and data analysis with minimal human prompting. It was compared to DeepSeek as a signal that Chinese AI talent could produce genuinely competitive frontier capabilities. Meta agreed to acquire it in December 2025, reportedly valuing the deal at between $2 billion and $3 billion, with plans to fold Manus's agent technology directly into Meta AI and accelerate Mark Zuckerberg's stated ambition to lead in the agentic AI space. The deal was structured to require a full exit from Chinese ownership and operations, with backers including Tencent and HongShan Capital receiving their distributions. By the time Beijing intervened, Manus engineers had already joined Meta's AI team, which is what makes the NDRC's order to unwind the transaction so disruptive. This is not a blocked deal in the ordinary sense. It is a retroactive veto on a transaction that had already substantively closed.
The legal basis Beijing used is worth understanding because it reveals the scope of the tool. The NDRC cited laws and regulations prohibiting foreign investment in the Manus project, without elaborating further. Chinese officials reportedly characterised the acquisition as a conspiratorial effort to undermine the country's technological foundation, which is language that places this far outside normal antitrust review. The review that preceded the decision encompassed export control regulations, foreign investment limitations, and competition law, and authorities at one point reportedly barred Manus employees from entering China during the process. This is not the kind of regulatory intervention that deal advisers can price into a timeline with legal analysis and patient negotiation. It is a national security framing that allows the government to block, delay, or unwind any transaction involving Chinese-origin AI talent or technology regardless of where the company is incorporated or how its ownership is structured at closing.
The implications for cross-border AI exit math are significant and immediate. Singapore had built a credible reputation as a neutral bridge for Chinese AI talent seeking access to international capital and, potentially, American acquirers. Bloomberg's May 2026 newsletter noted that the Manus decision dealt a blow to hopes that Singapore could serve as a sanctuary for Chinese AI, with the jurisdiction that was supposed to create regulatory distance from Beijing proving insufficient when the NDRC chose to assert authority anyway. For founders who have structured their companies in Singapore or other third-country jurisdictions specifically to position themselves for US acquirer interest, the Manus precedent removes a structural assumption that many cap tables have been built on. The company's incorporation address was always less relevant than the founding team's nationality, the origin of the IP, and whether Beijing would categorise the technology as strategic. Manus checked all three boxes in ways that ultimately mattered more than its Singapore headquarters.
For Meta, the damage is operational as well as financial. The company has already integrated Manus engineers into its AI team, and unwinding that integration is substantially more complex than reversing a financial transaction. Meta told both Bloomberg and TechCrunch that the transaction complied fully with applicable law and expressed confidence in an appropriate resolution, which is the language companies use when they have no good options. The agentic AI space is moving quickly, and Meta's investment thesis on Manus was specifically about accelerating capability development in autonomous agents as the primary competitive battleground after chatbots. If the deal is formally unwound and the engineering team has to be separated, Meta loses the integration benefit it paid for and its competitors gain time. OpenAI, Google DeepMind, and Anthropic are all developing agentic capabilities with domestic engineering teams that carry none of this geopolitical exposure, which starts to look like a structural advantage that was not visible until the Manus veto made it tangible.
The question for US tech companies considering AI acquisitions is whether they will now shift toward deal structures that eliminate Chinese-origin exposure entirely. The most likely adaptation is a preference for acquihires, minority investments in early-stage domestic companies, and full acquisitions of targets with no Chinese ownership, no Chinese-national founders in key roles, and no IP with traceable links to research conducted in China. That may sound like a narrow constraint, but it actually excludes a meaningful share of the AI startup ecosystem, given how much foundational research and engineering talent in the field has connections to Chinese universities, Chinese tech companies, or funding networks with Chinese limited partners. The Manus decision has effectively made that screening step mandatory for any American acquirer that wants to avoid an NDRC veto. For investors backing AI startups today, the founding team's passport and the origin of the training data are now diligence items with deal-closing implications, not just reputational footnotes. That is a material change in how cross-border AI investment works, and the full effect of it will show up in term sheets and cap table structures for years.
Also read: Infineon Just Proved the AI Boom Reaches Well Beyond GPUs and the Power Chip Layer Is Getting Repriced Too • Flex Is Spinning Out Its AI Data Center Unit and the Decision Shows How Deep the Infrastructure Premium Has Reached Into Industrial Supply Chains • Anthropic Just Launched 10 Financial Services Agents and FactSet Fell 8% on the Same Day for Good Reason
