Manus Is Not a Myth. It Just Hit the Boundary Earlier

On the surface, the story of Manus looks like a successful AI startup story: the product broke out, the team moved abroad, and the company was acquired by Meta. But look closer, and it stops looking like an ordinary startup myth. It reveals itself as a structural collision, one that surfaced early under the pressures of the new Cold War.

Let’s start with the conclusion: Manus’s success itself contained a strong element of contingency. That does not mean the company lacked ability. It clearly had product strength, and it did catch the agent wave. But it was not the kind of company that was “destined to win.” It was amplified because it happened to enter a narrative vacuum at the exact moment the market was hungry for AI agents. In March 2025, Reuters already placed Manus in the context of China’s AI surge and the growing attention around agent concepts. That alone shows that Manus’s breakout was not just a product event. It was also a narrative event.

What is really worth writing about is not that it became hot, but that the moment it did, it immediately exposed something deeper: this company was never operating in a truly safe middle ground.

Of course Manus originally wanted to start in China. The problem was that it simply could not gain real traction there. And very quickly, the issue stopped being just about market response. According to multiple people familiar with Manus’s early situation, police directly approached Manus and asked a strikingly blunt question: why are you letting Chinese people use overseas large models?

That sentence matters because it says the quiet part out loud. Manus was never just an ordinary startup project. The moment it tried to bring overseas frontier model capabilities more directly to Chinese users, it was bound to shift from being a “product” into becoming an “object” — something to be noticed, questioned, and redefined. From that point on, the company was no longer facing merely a growth problem. It was facing a control problem; no longer merely a business problem, but a sovereignty problem.

Many people analyzing Manus today wrongly mix it up with the path taken by the previous generation of Chinese internet companies.

The path that really existed in the past was not the Manus path. The old path was: build in China, serve the domestic market, list overseas, and cash out abroad.

The company’s center of gravity stayed in China. Its users were in China. Its cash flow was in China. Its growth was in China. “Going abroad” mostly meant a capital exit, not a shift in technological jurisdiction. The core assumption of that generation was simple: as long as the China market was large enough and the story was good enough, you could go to New York or another overseas market to raise money, list, and cash out. Didi became one of the clearest symbols of that old path breaking down. After going public in the U.S. in June 2021, it was hit within days by a cybersecurity review from the Cyberspace Administration of China, its apps were removed from stores, new user registration was halted, and it later moved toward delisting from the NYSE. The old formula did not disappear overnight, but its certainty began to break from that moment on.

Manus belonged to a different era, and it fell into a different trap.

It was not taking a mature China business and cashing it out in overseas capital markets. Instead, in the AI era, it was trying to stand on several things at once: • China’s engineering organization and talent base • America’s frontier models and open ecosystem • Third-jurisdiction buffers such as Singapore • A future exit into Western capital markets or M&A systems

In an earlier era, this might have been packaged as flexibility, cleverness, globalization. But in the AI era, under new Cold War conditions, this structure was conflicted from the start. If you depend on American model capabilities while burying your core organization deep in China, while also trying to use a third-jurisdiction structure to buffer the risk, and still hope to be cleanly acquired by a Western giant, that is not neutrality. It is placing yourself at the intersection of several opposing forces.

So when Manus later moved to Singapore, it cannot be described simply as “forced flight,” nor can it be described as “calm internationalization.”

A more accurate description is this: it was an outward move that was both voluntary and involuntary at the same time.

The voluntary side was that management had clearly realized that if the core organization remained entirely in China, its future room for business, financing, and exit would only keep shrinking. So it needed to move part of its people, structure, and jurisdiction outward as quickly as possible. The involuntary side was that this move was not carried out under conditions of full freedom or full calm. Because of EP limits and the reality of organizational separation, only part of the team could be taken out. The rest could only remain behind with fairly good compensation.

So Manus’s move to Singapore was not some elegant globalization upgrade. It looked more like a partly voluntary relocation carried out under the shadow of pressure: management wanted to get moving first, but the entire process of moving was already taking place under constraint.

Public reporting also supports this line. When Meta announced its acquisition of Manus in December 2025, Reuters described it as an AI company founded in China and later relocated to Singapore, with a deal size of around $2 billion.

But that deal did not resolve the problem. It raised it to a higher level. After being sold to Meta, events did not move into the standard ending of “startup success and graceful exit.” They moved rapidly into review. By March 2026, Reuters reported that, in the course of China’s review of Meta’s acquisition of Manus, two co-founders had already been barred from leaving the country. Of course, we cannot determine whether the two co-founders returned to China to undergo review because they had previously received signals or assurances from relevant parties.

That is what really makes Manus worth writing about. Its history isn't chaotic; it is remarkably self-contained.

First, it tried to get started in China; then it failed to gain traction; then, because the product itself touched a sovereignty-sensitive zone, it was directly questioned; then the organization began to move outward; then it was sold to Meta; then the transaction was reviewed; and finally, after the founders returned to China, they were restricted from leaving.

This is not bad luck. It is the sequential unfolding of a certain era’s constraints on startup organizations.

As for internal power structure, Manus was not the standard startup the outside world imagined. According to multiple people familiar with the situation, Ji Yichao was not the kind of operational founder who genuinely carried the company forward from start to finish. He looked more like a figure inserted into the structure by the capital side, someone who played a strong front-facing and mascot-like role in the overall narrative. This matters not because it is gossip, but because it reminds people of something important: in many companies like this, the people facing the public, the people facing capital, the people actually doing the work, and the people who truly hold key influence are not always the same group. The hierarchy visible from the outside is not necessarily the real control map inside.

And this is exactly where the biggest lesson of Manus for Chinese entrepreneurs lies.

First, the old path is dead. The formula of “build in China, serve China, list overseas, cash out abroad” stopped being a default viable channel after Didi.

Second, the new middle route in the AI era almost never truly existed from the beginning. This is not just about the environment getting worse later. It is that if you want to rely on American frontier models while keeping core R&D, core organization, and key personnel deeply in China, then use a third jurisdiction as a buffer, and finally enter the Western capital system, that combination is inherently unstable today.

Third, in the new Cold War, there is no room left for moderates. “Moderate” here does not mean emotionally moderate. It means a fantasy of organizational moderation: always believing you can avoid fully cutting ties, avoid fully leaving, avoid fully choosing, and keep all options open through ambiguity and delay — taking the benefits from every side first, and dealing with the consequences only after getting big enough. Today, this increasingly looks like self-deception.

Fourth, choosing sides is no longer a matter of attitude. It is a matter of organizational design. If from the start you intend to align with the U.S.-led side, then your company’s registration cannot remain in China, and your core R&D activity cannot remain in China either. It is not enough to move the shell abroad and call it done. From day one, legal jurisdiction, holding structure, core R&D, data routes, key infrastructure, and talent mobility paths all have to be designed around that reality. Otherwise you are not a global company. You are just delaying the real risk.

Fifth, in the AI era, the most sensitive thing is not the shell, but the R&D and control chain. Code, people, model access, data, and infrastructure are no longer just engineering problems. They are themselves part of the sovereignty problem.

So Manus should not be remembered as an AI startup story that “succeeded first and got unlucky later.” It is closer to a warning:

If you want to enter the U.S.-led open technology ecosystem, do not keep your vital organs in China.

In this era, moderation is not a strategy. Moderation is only delay. And delay, very often, is just another form of exposure.