The Day China Stopped Manus From Leaving — And What It Means for the AI World

On April 27, 2026, China drew a hard line around an AI company — and in doing so, may have defined the shape of global AI development for the decade ahead. China’s National Development and Reform Commission (NDRC) ordered Meta to abandon its planned $2 billion acquisition of Manus, a Chinese-founded AI startup that had taken the global developer community by storm since its launch in 2025. Meta had already integrated Manus into its internal systems following the December 2025 announcement. The NDRC’s statement was a few sentences long. The consequences will be far longer-lasting.

To understand why Beijing moved, it helps to understand what Manus actually is. Manus is not a narrow AI tool. It is a general-purpose AI agent capable of autonomously executing complex, multi-step tasks by combining tools in ways that earlier systems could not. When it launched, hundreds of thousands of developers joined a waitlist within days. The system’s ability to plan and execute across different domains — research, coding, content creation, web navigation — represented a qualitative leap beyond single-task AI assistants. Meta’s willingness to pay $2 billion reflected a straightforward judgment: Manus represented a decisive capability advantage in the race for AI agent dominance. That is precisely why Beijing could not allow it to leave.

The NDRC’s action was months in the making. Earlier this year, the Financial Times reported that Chinese authorities had barred two of Manus’s co-founders — Xiao Hong and Ji Yichao — from leaving the country while regulators investigated the transaction. The founders were prohibited from traveling abroad to pursue the sale of their own company. They could not board a plane to negotiate the future of what they had built. Meta described the transaction as having ‘complied fully with applicable law’ and said it anticipated ‘an appropriate resolution.’ There will likely be no appropriate resolution. NDRC rulings of this kind are not reversed.

The logic behind Beijing’s decision is not difficult to reconstruct. Since 2022, the United States has systematically restricted China’s access to advanced semiconductors, chip manufacturing equipment, and high-end GPUs. HBM memory, EUV lithography machines, the most capable AI accelerators — each category has been progressively restricted. The Manus acquisition represented the inverse scenario: a Chinese-origin technology flowing into the hands of the world’s largest American social media company. For Beijing, permitting that transfer would have been not just an economic loss but a strategic concession — an acknowledgment that China’s most valuable AI assets were for sale to the highest foreign bidder.

The semiconductor decoupling blocked hardware. The AI decoupling now reaches into software, models, and autonomous agents. Hardware can be stopped at a physical border. AI knowledge — algorithms, weights, training methodologies, architectural innovations — travels with people. You cannot easily stop it at customs. What China did instead was stop the people: by placing the founders under exit bans and ultimately prohibiting the transaction entirely. The instrument of control shifted from border inspection to personal restriction. That is a meaningful escalation in the toolkit of technology nationalism.

For Japan, this decision is not peripheral. Japanese companies operating across both US and Chinese technology ecosystems now face a sharper version of a question that has been building since 2022: when American and Chinese technology standards diverge, which do you choose? The automotive sector faces Chinese EV competition. The semiconductor supply chain is increasingly embedded in US-aligned networks. AI is now being governed by different rules in different markets. The Manus case shows that this is no longer a theoretical concern about future scenarios — it is a live constraint on actual business decisions, today.

The deeper structural shift is that China has now formally treated an AI company as state property. The NDRC’s prohibition means that Chinese-founded AI ventures — however privately organized and commercially motivated — can be prevented from transacting with foreign buyers if the state determines that the technology should stay inside China’s borders. This is a new category of national security intervention, applied to knowledge assets rather than physical ones.

There is something worth pausing on in the detail about the co-founders. Xiao Hong and Ji Yichao built something that attracted the attention of one of the most valuable companies in the world. The consequence of that success was that they lost the freedom to negotiate its future. The right to be compensated for what they created — to have someone recognize its value and pay for it — was taken away by the state. Whether those two individuals believe this outcome served them well is not something we will likely know for some time.

The AI development race is not slowing down. But the Manus decision signals that China intends to compete on its own terms, preserving domestic advantages and preventing cross-border technology arbitrage. The question now is whether any future Chinese AI breakthrough will be accessible to non-Chinese companies at all — or whether the next decade of AI simply develops in parallel, without a shared foundation. The global AI map is slowly splitting in two. Manus is one data point in that split. But data points have a way of accumulating into patterns, and this pattern has been building for years.