China's forced reversal of Meta's Manus acquisition has moved from court order to operational reality, and the cleanup is now the story every cross-border AI buyer should be watching.
Meta is now trying to separate itself from Manus, the Chinese-founded agentic AI startup it bought for more than $2 billion in December 2025, and the mechanics are proving far more complicated than a regulator simply saying no. According to Bloomberg, Meta has put a firewall between its systems and Manus's roughly 100-person team, cutting off staff access to internal data infrastructure as the company responds to Beijing's order to unwind the deal.
That is the important part. This is no longer a theoretical argument about whether national security reviews can reach into AI acquisitions after closing. It is a live test of what happens when engineers, code, product roadmaps, customer data, model behavior, and corporate reporting lines have already been joined together, then a government tells both sides to pull them apart.
The National Development and Reform Commission's April order was historic because it pushed beyond the normal pre-close review process. Meta had announced the Manus acquisition at the end of December, with the deal valued at more than $2 billion. Manus, founded by Chinese entrepreneurs and later based in Singapore, had become one of the most watched AI agent startups after its March 2025 launch. Its pitch was simple and attractive to a company like Meta: software that could execute multi-step tasks such as research, analysis, coding, website creation, and planning with less direct human prompting.
For Meta, Manus fit neatly into a broader AI race that has shifted from chatbots toward agents that can actually do things. The company has been trying to turn Meta AI into something more useful across WhatsApp, Instagram, Facebook, smart glasses, and business tools. Buying Manus gave it paying users, product momentum, and a team that had already shown it could ship quickly in one of the hottest corners of the AI market.
For Beijing, the problem was different. Manus may have moved its headquarters to Singapore and agreed to cut Chinese ownership ties, but Chinese regulators still viewed its technology history as relevant. The concern was not only who owned the cap table on closing day. It was whether AI technology developed partly in China could be transferred into the hands of a major American platform without further scrutiny.
The Hard Part Comes After Closing
Dealmakers like clean milestones. There is signing, approval, closing, integration. The Manus case shows how fragile that sequence can be when AI is involved. Once technical teams begin working together, the value of the acquisition is not sitting neatly in one folder that can be handed back. It is in shared architecture decisions, copied documents, product conversations, access permissions, model evaluations, customer learnings, and the human memory of engineers who have already spent months inside another company's systems.
That makes the firewall especially significant. In ordinary corporate language, a firewall sounds administrative. In this case, it is an admission that the cleanup has to be operational, not just legal. Meta has to show regulators that Manus personnel are no longer plugged into sensitive internal systems, while also protecting its own products from disruption and preserving whatever parts of the transaction remain legally defensible.
The risk for other acquirers is clear. AI deals are not like buying a small consumer app and waiting to migrate the user base. The integration often starts quickly because the strategic value is technical speed. If the asset is a model, an agent framework, a training workflow, or a specialized research team, the buyer wants that knowledge inside the company as soon as possible. That is exactly what makes a forced unwind so messy.
It also changes how boards should think about regulatory exposure. Pre-close approvals still matter, but they are no longer enough when the asset has a cross-border technology history. A company can be incorporated in Singapore, funded through offshore vehicles, and sold to a U.S. buyer, yet still trigger scrutiny if officials believe the underlying know-how was created under their jurisdiction. That is a much wider lens than many acquisition teams have been using.
A Warning For The AI Acquisition Boom
Meta is not the only company trying to buy scarce AI talent and product traction. OpenAI, Google, Microsoft, Amazon, and a long list of enterprise software companies are all competing for teams that can move faster than internal labs. Smaller startups know this, which is why agentic AI companies have been able to command rich valuations even before their business models fully mature.
The Manus situation will make some of those deals slower. Buyers will ask harder questions about where technology was developed, which employees had access to what code, whether government approvals are needed in more than one jurisdiction, and how integration should be staged before every review window is closed. That may feel cautious, but the alternative is now visible: a completed acquisition that has to be physically pulled apart after months of operational integration.
There is also a broader geopolitical message here. China is signaling that it wants a say in where sensitive AI capability ends up, even when the company has tried to rebase itself outside China. The United States has already been tightening its own review of outbound investment and foreign-linked technology transactions. The Manus case sits directly in the middle of that new reality, where AI products are commercial assets and strategic assets at the same time.
Also read: Alex Karp says every enterprise he works with is privately fed up with the frontier AI labs and warns the industry is sleepwalking toward nationalization • Texas Governor Greg Abbott just turned the AI data center boom into a bill that hyperscalers will have to pay themselves • China is building a legal wall around AI-displaced workers while America watches from the sidelines
The practical takeaway is not that AI acquisitions will stop. They will not. But the next wave of deals will need cleaner regulatory maps, slower integration plans, and a much clearer answer to one question: if a government orders this transaction unwound after closing, what exactly can still be separated?
