The Chinese-founded AI startup Manus has resumed operating as an independent company, completing the unwind of its acquisition by Meta that was ordered by Chinese regulators. The company announced the return on September 1, saying its founding team continues to lead the firm as an independent agent lab and that users can now restore data that had been deleted during the transition.
The reversal ends one of the stranger episodes in this AI cycle. Meta closed its acquisition of Manus on December 29, 2025, in a deal reportedly valued at about $2 billion. But the purchase quickly attracted scrutiny in Beijing, where officials worried about valuable AI technology and expertise leaving the country. In April 2026, China’s National Development and Reform Commission told the companies to separate.
How the unwind worked
The separation was executed in stages. Manus announced on August 11 that it would soon resume independent operation and laid out a transition plan for user data. Accounts created under Meta ownership faced a backup deadline of August 23, followed by a deletion window on August 23 and 24. Users could then begin restoring their data on August 25, after roughly two days with no account access. The company confirmed on September 1 that independent operations had formally resumed and that restores were live.
The company said in a blog post that its founding team would continue to lead the firm, and its roadmap points toward embedded workflows and proactive agents. Backers from its earlier rounds include Tencent, ZhenFund and HSG, the Chinese investors who backed the startup long before Meta entered the picture.
The whole unwind took a little over seven months from closing to reopening, faster than most contested mergers manage. That speed suggests both sides had an interest in a clean exit rather than a prolonged fight with the NDRC.
Why China blocked the deal
Manus built its name as an agent product, AI software that completes multi-step tasks on a user’s behalf rather than just answering questions. It drew heavy attention when it launched, and Meta’s acquisition was one of the largest deals for an agentic AI company. Chinese regulators saw the transfer of the technology to a US parent as exactly the kind of expertise outflow the country’s export-control and review regimes are meant to prevent.
The episode fits a pattern. China has tightened its grip on overseas AI transactions over the past two years, reviewing offshore acquisitions of domestic AI firms and restricting model technology exports. The NDRC’s order to unwind a completed acquisition, rather than block it in advance, was unusual and sent a signal that even signed deals can be reversed if they touch sensitive technology.
For global dealmakers, the case now sits alongside earlier interventions as a data point on Chinese review risk. Acquirers price regulatory risk in semiconductors and biotech routinely, but agentic AI software was largely treated as ordinary consumer tech. That assumption no longer holds.
What it means for users
For Manus customers, the practical effect is that the product continues but under original ownership. Some user data generated under Meta ownership was deleted as part of the separation, in line with the deletion windows the company announced in August. Reuters reported that the company said some user data will be deleted as part of its return to independence, with restore options for affected accounts.
For Meta, the unwind is a write-off of one of its larger AI acquisitions and a lesson in cross-border deal risk. The company had positioned Manus as a complement to its own AI assistant work. Neither firm has detailed the financial terms of the separation, including whether Meta recouped any of the purchase price.
There is also a hiring dimension. Employees who moved to Meta during ownership may face a choice about returning to the startup or staying with the acquirer. Manus’s statement that the founding team continues to lead suggests the core stayed together, but the engineering bench around it may take time to rebuild.
The wider agentic AI market
The independent Manus returns to a market that has crowded considerably. Agentic AI became the dominant product category of 2026, with OpenAI, Anthropic and Google all shipping agent products, and a wave of startups competing for developer and enterprise attention. Manus’s differentiation rests on its task-completion workflow and its positioning as a general-purpose agent rather than a coding or customer-service specialist.
The week’s other AI news added to a sense of overload. Anthropic, Meta, Google, OpenAI and an Abu Dhabi university all shipped new frontier models within four days, and Nvidia agreed to buy Hugging Face. In that environment, Manus’s return to independence is a niche story by volume, but a telling one by signal: the geopolitics of AI now reach into whether a startup can even change owners.
It also lands as China’s own AI companies turn to domestic capital. The New York Times reported this week that blockbuster market debuts by two Chinese technology firms show Beijing steering AI financing toward local investors and away from Wall Street. A startup that cannot be sold to a US buyer and cannot easily list abroad fits that system neatly.
Whether Manus thrives on its own is an open question. Independence means the founding team keeps control and Chinese investors keep their position, but it also means building enterprise trust without the balance sheet of a US parent, while competing against incumbents who ship weekly. The company’s next funding round, when it comes, will show whether investors price the independence as an asset or a constraint.
discussion