Tech advocates and analysts once imagined a world in which talent, capital, and ideas would flow freely between China and California. The blocked acquisition of Chinese artificial intelligence firm Manus by U.S. giant Meta is another moment when the fantasy of seamless globalization smashes into the hard wall of national security politics.
On April 27, the office coordinating China’s foreign investment security review mechanism, housed under the National Development and Reform Commission (NDRC), retroactively prohibited Meta’s acquisition of the Singapore-based Manus in December and ordered the parties to unwind the transaction. It’s a frustrating moment for the Chinese tech sector, despite it being used to taking hard blows from the government. Chinese AI start-ups have long aspired to “go global” to the United States. For AI companies, the global market, with its high-paying customers, looks more attractive than China’s crowded domestic battlefield. Global capital markets are deeper and more abundant.
There’s a well-known playbook for doing this, whether you call it “Singapore-washing,” “China-shedding,” or regulatory arbitrage: build the core capabilities in China; migrate the corporate shell to Singapore or another neutral jurisdiction; clean up the capitalization table and dismantle the Chinese base; and then sell to a U.S. tech giant or list offshore.
Manus seemed to embody the dream of taking a local success global. Launched in 2025 as a general-purpose AI agent, it was marketed as China’s answer to the era of agentic AI. It went viral almost instantly, with Benchmark leading a $75 million round. Meta later moved to acquire the company for a shocking $2 billion.
But the very magnitude of Meta’s offer, which made Manus’s success so dazzling, also did not bode well. The NDRC’s order to dismantle the deal has raised a series of questions.
Manus’s position in China’s AI ecosystem matters. Manus was not a foundation-model company in the DeepSeek mold, which makes the forced unwind especially curious. Unlike DeepSeek, Manus is a general-purpose agent whose core technology is not a self-developed large model. Instead, it orchestrates Gemini, ChatGPT, Claude, Qwen, search engines, tools, and retrieval systems to complete tasks.
In fact, Manus has long been labeled a “wrapper”—an app that coordinates other, more critical apps—which sits low on the technical prestige hierarchy. There is, arguably, relatively light “core tech” involved. Even founder Xiao Hong seemed to acknowledge this, arguing last year that “when a wrapper is done to perfection, it becomes genius,” and emphasizing customer experience as Manus’s competitive edge. Even in agentic AI, Manus does not appear to have a deep moat. OpenAI and Google are steadily building agentic capabilities into their own foundation models. Manus, in other words, does not necessarily have much irreplaceable core technology to speak of.
This is part of why Meta’s $2 billion proposal was so shocking. Did CEO Mark Zuckerberg overpay for a wrapper? More fundamentally, if Beijing’s concern were simply technology leakage, Manus is hardly the highest priority. And if media reports are accurate that Manus’s algorithms and model weights had already been handed over to the Meta team before Beijing acted, then any divestment at this stage is symbolic.
So why is Beijing acting so harshly?
First, Beijing needs to institutionalize its regulatory framework around agentic AI. Agents are potentially the next frontier after chatbots, the control layer for much more complicated tasks. Allowing a Chinese-born agentic AI company to be absorbed into one of the most powerful U.S. platforms would have set an undesirable precedent. A pipeline could emerge that dilutes China’s momentum in the agentic AI era, precisely where China has considerable comparative advantages: a vast domestic user market, abundant engineering talent, dense application ecosystems, and a brutally competitive environment that forces fast product iteration.
Beijing also needs to draw a brighter red line around Singapore-washing. Manus is not the first AI company to use such structures. Over the past decades, from the internet to mobile internet and now AI, many Chinese platform and technology companies have used this route: set up an offshore entity in the Cayman Islands or the British Virgin Islands—two common jurisdictions for corporate registration and overseas listings—as the vehicle, use a Hong Kong company to control a domestic “wholly foreign-owned enterprise,” and then have that enterprise sign contracts to control the onshore operating company. The essence of this variable interest entity (VIE) model is to replace equity control with contractual control, allowing Chinese companies to work around foreign-ownership restrictions, raise dollar funding, and ultimately list or cash out offshore.
In fact, the VIE structure was irreplaceable in Chinese technology history. In the early 2000s, foreign investment in the then-burgeoning internet sector was tightly restricted, and VIEs allowed Chinese firms to plug into the global capital system.
A warning shot was fired in 2011, when Jack Ma carved Alipay out of the VIE structure and converted it into a domestic company, citing the need to comply with Chinese central bank licensing requirements. But the transfer made to a Ma-controlled entity and disputed by Yahoo and SoftBank also exposed the vulnerability of the VIE model: Investors could have contractual claims to economic benefits without secure control over the underlying Chinese operating assets. Yet the route lives on.
So why the sudden scrutiny now? A decade ago, the regulatory logic was growth first. Today, both China and the world have changed. Politics and geopolitics have heightened security priorities. The legal ambiguity that once made VIEs a smart route is now fraught with pitfalls.
In the VIE structure, the legal operating entity sits onshore while effective control sits offshore. But then the hard questions begin. When users interact with the domestic operating entity and generate data, does control over that data belong to the onshore operating company or the Cayman parent? At what stage are security assessments required? In practice, personnel, data, and technical systems often become intertwined.
And it is not just tech leakage. The web of contracts underpinning VIEs—including informal equity pledges, exclusive service agreements, and voting-rights proxies—has always carried enforceability uncertainty, and Beijing cannot stomach such loopholes anymore.
The Manus halt’s shockwaves have already traveled through the broader VIE ecosystem. Dollar-fund limited partners are quickly reassessing the geopolitical risks and exit uncertainty of their Chinese technology investments. This will be bad news for small and midsized U.S.-listed Chinese companies. Some may even need emergency plans to dismantle VIEs—a process that can take years and cost millions of dollars.
Financing reality will likely keep the model alive. The depth of offshore capital markets remains unmatched, even as Hong Kong and domestic capital markets receive growing political support. For now, VIEs will remain important for Chinese firms seeking access to global capital.
But the Manus case does mean that there is a new, hard bargain facing Chinese entrepreneurs. They may have to choose between staying domestic and thus making do with thinner capital and a more limited market and starting abroad from day one, which comes with its own challenges. China’s domestic engineering system produces strong AI start-ups for a reason: abundant engineering talent, a dense ecosystem of iteration, and a domestic competition gym that hardens the best companies.
If the majority decide to establish themselves completely outside China altogether, that would be an ironic and damaging consequence for Beijing: In trying to prevent the loss of strategic technology, China may push the next generation of globally ambitious Chinese founders to avoid its jurisdiction earlier and more deliberately.
But Beijing appears willing to absorb that risk because the alternative looks worse: watching China’s most agile AI teams become feeder systems for U.S. platforms. Beijing seems to be betting that entrepreneurs will choose the alternative: staying domestic to avoid political risks.
Those risks have become particularly sharp. Just as the United States has tightened outbound investment rules and scrutinized U.S. capital flowing into Chinese AI, China is now asserting its own version of investment restrictions. This is another tool in China’s expanding legal and regulatory arsenal for technology competition with the United States, alongside export controls, supply chain rules, extraterritorial regulations, and more. Both sides are converging on the same conclusion: Security tools are now instruments of technological competition.
This is also why the official Chinese narrative around Manus is revealing. A well-known state-affiliated social media account has made the link between U.S. outbound investment scrutiny and Beijing’s investigation explicit: Manus first relied on domestic Chinese resources to incubate its development; then, under pressure from U.S. factors, tried to repackage itself as a Singaporean company; and finally sought to sell itself to foreign investors, thereby sidestepping Chinese regulatory requirements.
In that telling, the turning point came when Manus received Benchmark’s investment in April 2025. Roughly two weeks after the news was disclosed, U.S. authorities reportedly issued inquiries under their outbound investment security framework. A little over a month later, Manus restructured, cut China-based staff, and relocated its core team and headquarters to Singapore.
Beijing is drawing a direct line from Washington’s outbound investment controls to China’s own security review. This will become a new tool in China’s expanding arsenal of counter-leverage that Beijing can potentially deploy in future clashes with the United States, just as it did its control of critical minerals last year.
It is hard to be optimistic about Manus’s future. It is now deprived of the $2 billion exit that once seemed within reach, and the Chinese base it left behind has moved on. Agentic AI is no longer a novelty. After the failed sale, Manus risks becoming stranded between identities: dismissed by the Chinese public as an opportunistic betrayer and viewed by Beijing regulators as having engaged in regulatory gamesmanship.
To be fair, the jury is still out on whether Manus deliberately played with fire or whether its legal and auditing teams failed to properly assess the applicability of the NDRC security review requirements. AI companies already face overlapping constraints from China’s Data Security Law, Personal Information Protection Law (PIPL), generative AI regulations, and NDRC investment security review rules.
It is not clear how Manus’s lawyers anticipated the NDRC rule would apply or whether they simply assumed that the more standard data security, PIPL, and anti-monopoly requirements would suffice. After all, this marks the first publicly disclosed foreign acquisition in the AI sector to be directly blocked since the NDRC rule came into effect in the early 2020s. Manus’s legal team may simply not have had that on its radar.
But the NDRC rule could become the Damoclean sword hanging over China’s tech sector. It is extremely all-encompassing, casting a wide net around “security concerns.” It represents a penetrating, substance-over-form approach to regulation, regardless of whether a firm is Chinese on paper.
The old start-up mantra was “move fast and break things.” Manus moved so fast that it broke itself. Other firms may start to trace slower and more cautious routes.
