China Closes the Founders Exit

China’s National Development and Reform Commission this morning issued a brief statement ordering the cancellation of Meta Platforms’ two-billion-dollar acquisition of Manus, a Beijing-born agentic AI company that had already been redomiciled to Singapore in anticipation of the deal. The statement gave no reasoning. None was needed. For the past six months Beijing had grounded both co-founders of Manus, Xiao Hong and Ji Yichao, in a quiet exit ban; instructed peer companies including Moonshot AI and StepFun to refuse new American capital; and signalled, in language that anyone who had watched the Ant Group affair could read, that this acquisition was not going to complete. The cancellation is therefore the formal punctuation on a foregone conclusion. It is also, in a way that should command the attention of Western investors and policymakers, the most explicit statement yet of a doctrine that has been sitting in plain sight: in the People’s Republic, a founder does not own the company he builds.

The Pretext. The legal scaffolding for the block is technology export control. Beijing’s case, such as it is, rests on the claim that Manus’s underlying agentic stack was unlawfully transferred from Chinese soil to a Singapore parent in advance of the Meta sale, and that the transfer should have triggered a domestic export-control review that did not occur. This is a pretext, in the precise sense that it is procedurally adequate and substantively beside the point. The substantive question, the one Beijing actually answered, is whether a domestic AI capability of strategic value can be sold to an American buyer at any price. The answer is no, and the procedural anchor was found after the answer was determined.

What Manus Is. Manus is positioned as the first genuinely autonomous general-purpose AI agent, a system capable of decomposing a task, executing it across web tools, and returning results without human babysitting. It is built on a hybrid stack that fine-tunes Alibaba’s open-source Qwen model and routes through Anthropic’s Claude where appropriate. It does market research, books travel, codes, analyses stocks. The Western press has variously called it a second DeepSeek moment and an overhyped harness. Both descriptions are partly correct and beside the point. What matters is that Manus is the most successful Chinese-born agent product of 2025, with a footprint that extends into Western app stores, a valuation that crossed two billion before Meta tabled an offer, and a roadmap that places it inside the same product category Meta is most exposed on. For Beijing, that is not a startup. It is a capability.

The Bargain. Chinese founders do not raise capital, they are issued capital. The chain that funded Butterfly Effect, the Manus parent, runs on a permission infrastructure that begins long before any term sheet. ZhenFund, Sequoia China, Tencent: each is a private name on a public register, and each operates inside a regulatory perimeter the state can tighten or relax at will. Beijing does not appear on the cap table, but it is the silent first preferred share. It tolerates the company. It shelters the founder from the regulatory risks that flatten less favoured rivals. It permits him to fly to Riyadh and raise more. And it gathers in return a quiet expectation: when this firm matters, it will matter to us first. The price is rarely paid in the early years. It is paid at the exit.

The Lock-In. Exit bans on the co-founders of a private company are not an eccentricity of Chinese law, they are the bargain made visible. Once a startup crosses from private speculation into national asset, the founder loses the right to dispose of it. Liquidity becomes a state question. The pattern is consistent across regimes. Jack Ma’s Ant Group was eight days from a thirty-seven-billion-dollar IPO when the listing was suspended in November 2020, and Ma himself withdrew from public life. DiDi, which listed in New York against Beijing’s stated wishes in mid-2021, was driven from US markets within twelve months and forced into a Hong Kong relisting on terms set by the regulators. Bao Fan, the financier who built much of China’s modern technology investment-banking infrastructure, vanished from Renaissance Capital in February 2023 and resurfaced under official custody. The mechanism varies; the principle does not. A founder who attempts to exit a national capability on terms unfavourable to the state will find that the state has anticipated him.

The Western Mirror. Silicon Valley is built on the right of exit. The strategic acquisition, the IPO, the secondary, the founder’s modest second house: these are not corruptions of the model, they are the model. American founders take state-adjacent money, including DARPA grants, In-Q-Tel positions, and the broader cluster of defence-tech contracts that animate the latest crop of unicorns, and remain free to sell tomorrow. The Chinese founder takes private capital and is not. Same balance sheet, opposite property right. The cap table tells a Western analyst nothing useful about who actually owns the company. The political register tells him everything. This asymmetry has been latent in Sino-American capital flows for at least a decade, but the Manus block is the moment it becomes operational policy.

The Doctrine. Xi-era industrial policy has, more or less openly, redefined what a Chinese tech company is for. It is not a vehicle for shareholder return. It is a unit of national capability, instrumented to the long-run technological autonomy of the state. The founder is the steward of that capability, not its owner. Once that premise is accepted, and Beijing has been unsubtle about asking founders to accept it, the Manus cancellation is not protectionism in the ordinary sense. It is the system enforcing its own internal logic. A national capability cannot be sold to the country it was built to compete with. Of course the deal dies. Anything else would be the system contradicting itself in public.

The Implications. For Western capital and policymakers, the operating consequences are sharper than the rhetoric suggests. The first is that the strategic acquisition vector for Chinese AI capability is now substantively closed. Meta’s two billion was the test case, and the test failed publicly enough that no general counsel will recommend a similar transaction within the foreseeable horizon. The second is that any private holding in a Chinese AI firm of strategic interest must now be modelled with a state-veto premium attached to liquidity. Western limited partners exposed via Sequoia China-vintage funds, Tencent-adjacent vehicles or Hong Kong secondaries should expect significant haircuts on terminal value. The third, and most uncomfortable, is symmetric. If Beijing has now publicly asserted a sovereign property right over its founders, the corresponding policy answer in Washington, Brussels and London is not to reciprocate (Western states should not become creditors of last resort over their entrepreneurs), but to treat Chinese outbound capital with the same political register Beijing applies to American inbound. CFIUS exists; its EU and UK equivalents exist; their willingness to enforce will be the only thing that prevents the asymmetry from compounding.

There is no scandal in any of this, and that is the part worth sitting with. The Chinese state has not done anything it did not say it would do. The cage has been sitting in plain sight for the better part of a decade, and a generation of founders chose to walk into it because the capital, the protection and the runway were too generous to refuse. Meta’s two billion will not land in Xiao Hong’s account. It will not land anywhere. The asset he built is a permanent ward of the state, and so, in a quieter and less brutal sense, is he. That is the trade. The state funds you, the state keeps you. The only way out is not to have walked in, and that lesson, finally, is the one the next cohort of Chinese AI founders will have to study.

Read our full Report Disclaimer.