Commentary: How China manages Singapore-based Manus’ sale will reveal its AI ‘red line’

WASHINGTON DC: When Meta moved to acquire Butterfly Effect, the Chinese-founded startup behind the AI assistant app Manus, it looked at first like a neat success story: Build in China, move the holding company to Singapore, then sell to a US tech giant for US$2 billion.

Then on Jan 8, China changed the script: The deal would first face a formal review of whether this violates tech export or national security rules.

Manus went viral in tech circles last year, claiming to be the world’s first general AI agent that could make decisions and handle complex tasks with minimal human guidance. Influential tech figures like Twitter’s Jack Dorsey and AI platform Hugging Face’s Victor Mustar sang its praises.

But from Beijing’s perspective, it is less about the potential major breakthrough in agentic AI, and more about what the fate of Manus might signal. Recent reporting in the Financial Times suggests that this is exactly how China’s leaders see it.

The review was reportedly triggered after top officials warned against “selling young crops” – shorthand for letting emerging technologies and talent be sold offshore too early – and asked the commerce ministry to assess whether the Meta-Manus deal would lead to a loss of cutting-edge capabilities. Venture investors now see the case as a test of what kinds of exits Beijing will tolerate for Chinese-founded AI firms.

NOT JUST ANOTHER APP DEAL

Manus lands in an awkward place. It is a consumer app from a firm already re-domiciled to Singapore. On paper, there is limited scope for a narrow national security case against the sale.

Yet the commerce ministry is both the gatekeeper for trade and export controls and China’s main representative in trade talks with the United States. It has little desire to be seen as jeopardising the fragile trade truce agreed in October 2025, just as Presidents Xi Jinping and Donald Trump are preparing to meet repeatedly in 2026 to keep that understanding on track.

At the same time, it is unlikely to simply stand aside while a Chinese-founded AI company of global interest is folded into a US tech giant.

The commerce ministry cannot veto the sale to Meta outright. It would be hard to defend on technical grounds and risk embarrassing the US Treasury.

But doing nothing is also untenable. Other tech companies might get the idea that clever corporate structuring is enough to escape Beijing’s reach.

A formal review sits uneasily in the middle, but it does three things the commerce ministry needs: It asserts its regulatory presence, avoids an immediate clash over the truce, and passes the hardest choices up to the top leadership.

THE ECOSYSTEM THAT PRODUCED TIKTOK

Manus is unlikely to be viewed as a narrow trade case by top leaders. Instead, it will bring into focus how China’s tech economy actually operates.

Despite a growing role in open-source foundation models, China’s comparative strengths still arguably lie at the application layer.

TikTok is the clearest illustration: a product first stress-tested on China’s vast home market and then turned into one of the world’s most successful consumer apps. The ecosystem that made that possible – engineers, product managers and operators who can ship, iterate and scale quickly – can still be seen as one of China’s core assets.

The long-running wrangle over TikTok’s forced divestment in the US, which has now produced a majority American-owned joint venture while the app’s Chinese parent company ByteDance retains only a 19.9 per cent stake, shows how an application-layer success can become a bargaining chip in negotiations with Washington.

NOT STRATEGIC ENOUGH?

Industrial policy signals, however, pull in a different direction. Under the banners of “new quality productive forces” and technology self-reliance, China’s state capital is increasingly being directed towards what are seen as strategic infrastructure-type projects.

Those are not consumer apps, but the next DeepSeek-style frontier model, the next Nvidia and other large-scale AI platforms.

The National Development and Reform Commission, China’s central AI policy planner, has been at the centre of that push. According to the FT report, it had initially declined to intervene when Manus first moved to Singapore, apparently judging that an AI agent app did not yet fall into that strategic infrastructure category.

Many regions and sectors now talk about building their own large-model systems – in mining, logistics, finance, medicine and education – some of which are already being presented as early success stories.

It is relatively easy to frame such projects as aligning with Beijing’s priorities. It is much harder to explain why supporting one of many app-layer start-ups serves the national AI interest.

Mr Xi has already cautioned against this kind of copy-and-paste planning in December 2025, criticising provinces that “ignore their own conditions and blindly chase the latest craze” and asking why “everyone is trying to do exactly the same thing”.

That warning maps closely onto local fiscal realities. Many governments that have signed up to the AI push still face shrinking land-sale revenues and hidden debts, and most struggle to fund world-class foundation models when hardware and top talent are scarce and expensive.

In that context, application-layer bets are often more fiscally realistic than frontier-model projects. The problem is that they are not particularly lucrative at home when there is limited willingness to pay for consumer AI apps and little access to public sector contracts.

AI-based agents and productivity apps are concrete proof that a locality is doing something in AI for alignment with Beijing, but they often need access to global users and foreign customers to justify the upfront investment.

THREE SCENARIOS OF HOW CHINA HANDLES MANUS

That is why the signal sent by the Manus review matters. Other successful app-layer firms will be watching to see if they can legitimately look for international consumers.

The overlapping objectives – protect the trade truce, show regulators are in control and preserve China’s strengths in the app-level innovation – mean that how Manus is handled will reveal Beijing’s policy preference.

The main options are relatively clear, and each would tell a different story about what now matters most to Beijing.

First, a relatively quick approval. It would be read as a sign that preserving the trade truce and being realistic about China’s tech advantages matter more than making an example of a non-strategic firm. It would suggest that the main strategic concerns still lie higher up the AI stack and that Beijing is, for now, prepared to tolerate some internationalisation of application-layer companies built on top of that infrastructure.

Second, a conditional approval paired with visible limits on what code or data can be transferred. It would give more weight to the commerce ministry’s desire to be seen as a serious regulator of outbound AI without tearing up the diplomatic script. It would hint at a more layered approach to governance: treating core models and compute as tightly bound into the national security lens, while allowing some app-layer firms to internationalise under defined terms.

Finally, a prolonged procedural limbo. This would suggest that no single calculation has clearly prevailed. Stretching the review out would preserve room for diplomatic manoeuvre in future US-China trade talks and deter copycat restructurings.

It would also buy time for senior leaders who are still working out, case by case, where China’s real AI advantages lie. The commerce ministry would have more space to establish itself as a serious voice on AI policy domestically. For these reasons, delay may be the path of least resistance, politically safer than a clean “yes” or “no”.

But at what cost? Founders, local governments and foreign investors will think twice about backing similar products if they cannot be allowed to go global.

Wang Shengyu is a research assistant at the Asia Society Policy Institute’s Center for China Analysis.