China is increasingly regarding AI technology as a national treasure.

Yesterday, the National Development and Reform Commission (NDRC) issued a terse statement barring foreign entities from acquiring the Manus project and demanding the transaction be rescinded, offering no further explanation.

When Meta acquired Manus for $2 billion at the end of last year, many people immediately expressed congratulations. A Chinese AI startup being acquired by the world's largest social media company, with its founder becoming a Meta vice president—the rapid ascent from zero to a $2 billion valuation in just 12 months was widely hailed as the ultimate success story for a Chinese AI startup. Now, the NDRC has declared this deal null and void and mandated its reversal.

Why is the sale not permitted? The announcement did not provide details, but the answer lies in an emerging concept: sovereign AI.

I. AI Transcends Mere Business

In recent years, within China's policy framework, AI has been primarily viewed as an 'industry.' Terms such as 'AI industry,' 'AI+ initiatives,' and 'smart economy' suggest that AI is a commercial endeavor that requires scaling up. Companies develop technologies, products, and commercialization strategies, while the government provides policy and financial support, allowing the market to determine success or failure.

However, the Manus case represents a turning point: AI in China is no longer just an industrial concern but has evolved into a matter of national sovereignty.

What is sovereign AI? Simply put, it means that a nation's core AI capabilities must remain under its own control, rather than being dominated by foreign entities or transferred to rival nations. Just as there are export controls on chips and rare earths, AI technology is now subject to similar restrictions. The U.S. uses chip bans to constrain China, while China employs security reviews to prevent its AI technology from being acquired by the U.S. Both sides are essentially treating critical technologies as national assets that must be safeguarded.

What is Manus's technology? It is the world's first general-purpose AI agent, which quickly gained widespread attention upon its launch, achieving $100 million in annual recurring revenue (ARR) within a year. It can autonomously operate browsers, read and write files, send and receive emails, and execute complex tasks. This is not merely a chatbot—it is an AI that can act on your behalf. If Meta had acquired and integrated this capability into its ecosystem, China would have permanently lost this technological edge.

Examining Manus's timeline over the past year reveals just how highly sought after this technology is. Launched in March last year, its website received over 10 million visits within four hours, with invitation codes being resold for tens of thousands of dollars. A month later, it secured $75 million in funding at a $500 million valuation. By year-end, its ARR exceeded $100 million, and Meta offered $2 billion, finalizing the deal in just over ten days of negotiations. From obscurity to acquisition by a global tech giant in 12 months—this rapid progression indicates that leading companies in the AI agent sector have become focal points in the tech competition between major powers. This is no ordinary business.

II. The Boundary Set by the NDRC

The transaction structure for Manus included numerous 'avoidance' tactics. The founding team first relocated the company from Wuhan to Singapore, changing the operating entity to a Singaporean company and moving the core team offshore. Then, as a Singaporean company, it was acquired by a U.S. firm. Formally, this appeared to be a transaction unrelated to China.

However, the NDRC looks beyond formalities to the substance. The code was written in China, the models were trained in China, the data was accumulated in China, and the core engineers were Chinese. Changing the registration location does not alter the technology's 'nationality.' Once this logical chain is established, all attempts to transfer AI technology through paths like 'domestic R&D + offshore shell + foreign acquisition' are blocked.

The line drawn by the NDRC is clear: AI technology developed within China cannot have its control transferred overseas without review. Regardless of whether you're registered in Singapore or the Cayman Islands, or how your operating entity changes, if the core technology originated in China, transferring control requires security review. This line was previously ambiguous but has now been clarified by the Manus case.

However, drawing the line is easier than enforcing it. Manus is not a factory—you cannot simply return the keys to restore the status quo. Its most valuable assets—code, models, and engineering expertise—exist in people's minds and on servers. After Meta acquired Manus, it integrated the technology into its advertising system and even launched new applications. The 40 core engineers are already working in Meta's Singapore and U.S. offices, with some applying for green cards. Equity can be returned, and money can be refunded, but how do you retrieve the knowledge in an engineer's mind? How does code already used by Meta count as 'restoring the status quo'? Reversing this transaction may be even more challenging than completing it initially.

III. It's Not Just About Manus

Viewing this solely as a halted transaction misses the bigger picture.

For years, the core logic of global tech competition has been: the U.S. restricts chips, and China restricts rare earths. Both are 'hardware' strategic resources. But AI is different—its strategic value lies in software, algorithms, data, and engineering expertise. Unlike chips, which can be physically intercepted, these assets are hidden in code, people's minds, and servers. How do you prevent an algorithm from being 'taken'? How do you stop an engineer from bringing technical expertise to another company?

The Manus case is China's response to this question: regulate from the source. Regardless of where the technology is now, if it was developed in China, its transfer requires review. This logic mirrors U.S. chip export controls. The U.S. says Nvidia's chips can't be sold to China; China says AI developed by Chinese teams can't be sold to the U.S. Both sides are taking another step toward 'technological decoupling' in AI.

For Chinese AI startups, this means rethinking exit strategies. Previously, the ideal exit for Chinese AI investments was a high-price acquisition by overseas giants like Google, Meta, or Microsoft—a win-win scenario. After Manus, the risks of this path can no longer be ignored. In the future, exits for Chinese AI startups will rely more on domestic listings or acquisitions by Chinese companies rather than overseas buyers.

For individual entrepreneurs, the impact is more direct. Previously, the exit path for an AI company might have involved scaling user numbers, securing multiple funding rounds, and finally being acquired by a major player, domestic or foreign. Now, foreign acquisitions face an additional security review hurdle. From signing to being halted, the Manus deal took only four months—during which the team had relocated, engineers had joined Meta, and partial payment had been made, making reversal extremely costly. Any entrepreneur considering foreign acquisition must now ask: Will this transaction pass security review? This question didn't exist before but is now mandatory.

IV. A Broader Trend

In the larger context, Manus is just one wave in the 'sovereign AI' tide.

Globally, more countries are treating AI as a strategic asset. The EU is pushing its AI Act, requiring high-risk AI systems to undergo review within the EU. India mandates that tech companies store Indian users' data within the country. Middle Eastern sovereign funds are heavily investing in AI companies—not for financial returns but to avoid marginalization in the AI era.

AI is transforming from a 'globalized technology' into a 'sovereignized asset.' Just as oil transitioned from a commodity to a national strategic resource in the 20th century, AI is following the same path in the 21st. Whoever controls core AI capabilities—models, computing power, and data—holds the trump card in future national competition.

You might think 'sovereign AI' is distant from your life. But consider: if you work at an AI company, the value of your stock options and the company's exit strategy depend on this line. If you invest in AI projects, your exit options are narrower, and valuation models must be recalculated. Even as an ordinary user, as the global AI ecosystem shifts from 'one network' to 'multiple networks,' the AI products and services available to you will change accordingly.

The NDRC's brief announcement may seem small, but it signifies something much larger: China has decided that AI is a national asset, not a freely tradable commodity. Manus is the first to hit this line, but it won't be the last.