Meta-Manus Acquisition: $2B Price & China's Unwind Order

What Meta actually bought

Manus began inside Butterfly Effect, a company Xiao Hong founded in Beijing in 2022. Butterfly Effect's first product, released in 2023, was Monica: a browser extension that put several vendors' language models behind one sidebar for translation, summarizing and writing help. Manus was a different kind of product, an agent meant to take an instruction and go do the work, browsing, writing code, producing a finished file, rather than answer inside a chat window. [4][5]

The invite-only preview went up on March 6, 2025, and access codes started reselling on Chinese secondhand marketplaces for the equivalent of thousands of dollars within days. Butterfly Effect raised a $75 million Series B that April, led by Benchmark, at a reported valuation of about $500 million. Earlier rounds had come from ZhenFund, and from Sequoia China (renamed HSG) and Tencent. [5]

By June 2025 the company had moved its registered headquarters to Singapore, incorporating as Butterfly Effect Pte. Ltd. Roughly 40 of its 120 China-based employees relocated; the rest were laid off. What relocated with them was growth fast enough to interest a buyer: Manus said on December 17, 2025, that it had crossed $100 million in annualized recurring revenue eight months after launch, with another roughly $25 million in usage-based revenue on top, and that it had been growing more than 20 percent a month since a mid-2025 product update. [5][6]

The price and the condition attached to it

Meta and Manus announced the deal on December 29, 2025, without disclosing a specific price or a cash-versus-stock split; reporting put the value above $2 billion. Meta absorbed Manus's roughly 100 employees and folded the product into its own push toward agents that complete tasks unsupervised, the same pitch, market research, coding, data analysis, Manus had already been selling to its own customers. [2][1][3]

Meta's stated condition was no continuing Chinese ownership interest in Manus after the deal closed, plus a wind-down of Manus's operations in China. CEO Xiao Hong framed the sale to press as continuity rather than change: Manus would keep running its app and website independently, he said, on "a stronger, more sustainable foundation without changing how Manus works." [1][2]

Atlas interpretation: The condition was built around a fact already true on paper: Manus's holding company was Singapore-incorporated, its declared headquarters was Singapore, and whatever Chinese equity remained in the cap table could be cleared out as part of the sale. What that condition did not reach was the part of the company still tied to China regardless of where the holding entity sat, namely the two founders, both Chinese nationals, and the technology and much of the surviving workforce that had built the product before the move. [1][5]

The review nobody announced starting

The instrument was Order No. 37, the Foreign Investment Security Review Measures the NDRC and China's Ministry of Commerce had jointly issued in 2021. On April 27, 2026, four months after the deal was announced, the NDRC's Office of the Working Mechanism ended the review it had been running with a short statement: the commission had made a decision "to prohibit foreign investment in the Manus project in accordance with laws and regulations," and required the parties involved "to withdraw the acquisition transaction." [9][7]

Lawyers who track the measure called it the first publicly disclosed case of an AI-sector deal blocked under Order No. 37 since it took effect in 2021. The order did not ask Meta to renegotiate anything. Article 12 of the measure allows an outright prohibition of a transaction judged to touch national security through key technology or critical infrastructure, and the decision instructed the parties to "take necessary measures to restore the pre-investment state," months after the acquisition had already closed and Manus staff and systems had begun moving onto Meta's infrastructure. [9]

Around the same time, Chinese authorities barred Manus's two cofounders, Xiao Hong and Ji Yichao, from leaving the country, a step Fortune described as routine once officials open this kind of investigation. Meta's public response stayed short: the transaction "complied fully with applicable law," the company said, and it expected "an appropriate resolution to the inquiry." [8][7]

What the offshore move didn't fix

Atlas interpretation: Meta's ownership condition treated a clean cap table as the whole problem: strip out the Chinese investors, and a Singapore-incorporated company with a Singapore address is a Singapore company. Order No. 37 does not work that way. It reaches transactions that touch national security through the technology and infrastructure involved, not only the nationality on a share register, and Manus's technology, its founding team, and much of its remaining workforce were still Chinese in origin even after the 2025 move. Relocating 40 of 120 employees to Singapore changed where the office badges said the company was based. It did not change where the agent had been built, or by whom. [9][1][5]

Atlas interpretation: That is why the block reads as a marker rather than a one-off. Washington had spent years restricting American investment into China's AI sector and export of the chips that train it; the Manus order is Beijing doing the mirror-image thing, screening outbound technology and talent the way the United States screens outbound capital and inbound chips. A Chinese-founded team relocating its headquarters and raising Western venture money no longer settles, on its own, which government gets a say in whether the company can be sold. [8][7]

Meta's year of buying its way into agents

Atlas interpretation: Manus capped a run of 2025 purchases meant to compensate for Meta's own AI research falling behind, after Meta's June 2025 investment in Scale AI put $14.3 billion into 49 percent of Scale AI and its founder, brought in to run a new superintelligence lab. Both deals shared the same premise: that a research organization assembled by acquisition would produce results faster than one Meta grew from inside. [3]

Atlas interpretation: The pattern held even as the people around it churned. Yann LeCun announced he would leave the lab that Scale AI money had funded six weeks before the Manus deal was announced, after years of arguing publicly that scaling language models would not reach human-level intelligence. Buying an agent company built on exactly that scaling bet, six weeks after its own chief AI scientist announced his departure over the argument, was less a rebuttal of him than a sign Meta had stopped waiting for the debate to resolve and started buying whatever was already making money. [3]

Atlas interpretation: That willingness to buy first and settle ownership questions later ended up costing Meta the one deal in the run that involved a Chinese-founded team. China's NDRC blocking the deal four months after the acquisition closed showed that in AI, a founding team's origin can outlast a company's own relocation, incorporation, and press release. [1][7]