Beijing Presses Rewind: Tencent Leads a $2 Billion Buyback of Meta's Blocked Manus Deal

An Acquisition Running in Reverse

Acquisitions fail all the time. They fall apart in diligence, they die over price, they get abandoned when the market turns. What almost never happens is the thing now unfolding around Manus: a deal that closed, that operated, and that a government has ordered its participants to undo.

According to reporting first published by the Financial Times and subsequently picked up by Reuters and Bloomberg, Tencent is in talks to become the largest shareholder of the AI agent startup Manus, leading a consortium that would buy the company back from Meta. The consortium reportedly includes Manus's early backers ZhenFund and HSG — the firm formerly known as Sequoia Capital China — and the repurchase price is described as no less than $2 billion, the same figure Meta agreed to pay.

The reason a buyback is necessary at all is the part worth sitting with. Meta announced the acquisition in December 2025, part of Mark Zuckerberg's broader push to buy his way into agentic AI. In April 2026, Chinese regulators opened a review and ultimately ordered the transaction unwound, citing potential violations of investment rules. Beijing did not block a pending deal. It reversed a completed one.

What Manus Actually Is

Manus is not a household name in the way that ChatGPT or Gemini are, and understanding why anyone would fight over it requires knowing what it does. The company built what it marketed as the world's first general AI agent — software that plans, calls tools, gathers information, and executes multi-step tasks with minimal human supervision. It is squarely in the category that every major lab has spent the past two years chasing: not a chatbot that answers, but a system that acts.

That is precisely the category Zuckerberg was shopping for. Meta's agentic ambitions have been well-telegraphed, and buying a working agent company with real users is faster than building one. It is also, evidently, more politically fraught than Meta anticipated.

The commercial case appears to have held up during Meta's brief ownership. Reports citing Chinese-language sources suggest Manus's annualized revenue climbed to somewhere between $400 million and $500 million, up from roughly $100 million before the acquisition. Treat those numbers with real caution — they are secondhand, unaudited, and unconfirmed by any of the parties. But if they are even directionally right, they explain the buyback price: this is not a distressed asset being dumped. It is a growing business that Beijing decided should not be American.

A Singapore Address Was Not Enough

The detail that deserves the most attention is one that is easy to skim past. Manus had already moved its operations from China to Singapore the previous year, a relocation widely understood as a bid for international operational flexibility — a way to serve global customers, take global capital, and sit outside the gravity well of Chinese tech regulation.

It did not work. The company's Chinese roots still triggered a regulatory review, and that review still produced an order to unwind. For every Chinese-founded AI startup that has quietly re-domiciled in Singapore, Delaware, or the Cayman Islands on the theory that a new address buys jurisdictional distance, this is a discouraging data point. Beijing appears to have taken the view that where a company's founders and technology came from matters more than where its holding company is registered.

That principle, if it holds, has a long reach. The Singapore-registered AI startup with a Chinese founding team is not an edge case. It is one of the most common structures in the sector.

Why Tencent, and Why Only a Minority

The proposed ownership structure is the tell. Reporting indicates Tencent would hold the largest single stake while remaining a minority shareholder — an arrangement that, as one account put it, keeps any one investor from controlling a startup whose Chinese roots proved politically delicate.

Read that carefully. The structure is engineered to avoid the appearance of any single owner, including Tencent, exercising control. Beijing does not appear to want Manus owned by an American giant. But it also does not obviously want it swallowed whole by a Chinese one, at least not in a way that invites the next round of scrutiny — from Chinese antitrust regulators, from foreign governments, or from customers outside China who would have to decide whether to keep using an agent platform controlled outright by Tencent.

The composition of the buying group tells the same story. Early backer Benchmark — the American venture firm — is expected to sit out. The cap table is being converted from mixed Chinese-American ownership into something predominantly domestic. That is not an accident of who happened to want in. That is the point of the exercise.

Hype Versus Real

It is worth being precise about what is confirmed and what is not, because the gap is wide.

Confirmed and consistently reported: Meta announced the $2 billion acquisition in December 2025; Chinese regulators reviewed the deal in April 2026 and ordered it unwound; Tencent is in talks to lead a buyback consortium at a price of no less than $2 billion.

Not confirmed: the final terms, the timeline, the precise stake sizes, and whether the deal closes at all. When Reuters sought comment, none of the parties — not Tencent, not Manus, not Meta, not ZhenFund, not HSG — immediately responded. This is a story built on sources, not on filings or press releases. Talks that are reported as ongoing sometimes end in nothing.

The revenue figures are the weakest link in the chain and should carry the largest asterisk. A jump from roughly $100 million to $400–500 million in annualized revenue would be extraordinary growth, and it is being relayed through Chinese-language reporting that no principal has verified.

So the hype to resist is the narrative that Manus is a proven half-billion-dollar agent business that Beijing rescued from foreign hands. The reality to take seriously is narrower and, honestly, more important than the revenue: a sovereign government demonstrated that it can reach into a closed cross-border AI transaction and reverse it, and the market is now organizing itself around that fact.

What This Does to Cross-Border AI Deals

For the past several years, the implicit assumption behind Western acquisitions of Chinese-founded AI companies was that regulatory risk was front-loaded. You cleared the reviews, you closed, and then the asset was yours. Manus breaks that assumption. Closing is no longer the end of political risk; it may just be the beginning of it.

The practical consequences follow quickly. Deal prices for Chinese-founded AI assets should carry a discount for reversal risk. Acquirers will want indemnities that are difficult to write and harder to enforce. And boards at companies like Meta will have to ask a question they could previously wave away: what happens if we buy this, integrate it, staff it, ship it — and are then told to give it back?

There is a symmetry here that should not be lost. Washington has spent years restricting what American technology can flow to China, through export controls on chips and scrutiny of Chinese investment in US tech. Beijing has now shown it can run the same play in the opposite direction, blocking the flow of Chinese AI capability into American ownership. Agentic AI is being treated by both capitals as strategic infrastructure, not as software.

The Takeaway

The Manus unwinding is a small deal by the standards of a year in which AI infrastructure commitments are measured in tens of billions. Two billion dollars barely registers next to the data center leases and memory financings dominating the headlines. Its significance is not its size. It is the precedent.

Beijing has established that a completed acquisition of a Chinese-founded AI company is reversible, that a Singapore relocation does not confer immunity, and that the replacement ownership structure will be designed to its satisfaction — domestic-led, but with no single controlling hand. Tencent gets the largest stake precisely because it is not allowed to have all of it.

Watch whether the deal actually closes, and watch the price. If the consortium pays the full $2 billion, Meta exits whole and the story reads as a managed, orderly reversal. If it closes materially lower, the message to every foreign acquirer eyeing a Chinese AI asset is blunter: you can buy it, but you may not get to keep it, and you may not get your money back either.

Sources: Reuters · The Next Web · TechStartups · Bloomberg