China Just Forced Meta to Unwind a $2 Billion AI Deal — And It Should Worry Every Facebook User

By Lynn Matthews, WECU Media

For months, Meta executives believed they had pulled off one of the most important acquisitions in the company's history: a roughly $2 billion purchase of Manus, one of the world's most talked-about autonomous AI agents. The deal closed quietly in December 2025. Engineers were absorbed. Code was folded into Meta's systems. The technology even began showing up inside Meta's advertising tools, the same machinery that helps generate the bulk of the company's revenue from its nearly 3.4 billion daily users across Facebook, Instagram, WhatsApp and Messenger.

Then, in late April 2026, Beijing said no.

China's National Development and Reform Commission — the powerful state planning body that functions as the country's gatekeeper for sensitive cross-border deals — ordered Meta and Manus to tear the transaction apart entirely, according to TechCrunch and the Wall Street Journal. No detailed explanation was offered. The order simply directed "the parties involved" to withdraw the transaction "in accordance with laws and regulations."

For a company that built its empire on the idea that the internet — and the data, code and talent that move through it — doesn't really belong to any one country, the message from Beijing was blunt: that idea no longer applies to artificial intelligence.

A Deal Nine Months in the Making

Manus first became a global sensation in early 2025, when a small Chinese startup called Butterfly Effect released a preview of an AI agent that didn't just answer questions — it acted. Manus could browse the web, write and execute code, build websites, plan trips and complete multi-step research tasks largely on its own, without a human walking it through each step, according to the South China Morning Post. Within weeks, it was being compared favorably to research tools built by OpenAI.

The company behind it had deep roots in China. Butterfly Effect was founded in 2022 by Xiao Hong — known as "Red" — a Wuhan native and graduate of Huazhong University of Science and Technology, according to Wikipedia's entry on Manus and reporting from ChinaTalk. Co-founders Ji Yichao and Zhang Tao rounded out the founding team. The company operated out of Wuhan and Beijing, and its earliest product wasn't Manus at all — it was Monica, a browser plug-in that let users tap into ChatGPT, Claude, Gemini and other AI models from a single sidebar.

The early money came almost entirely from Chinese investors: ZhenFund, Tencent and HongShan Capital Group (formerly known as Sequoia China) all backed the company through multiple funding rounds, per 36Kr and ChinaTalk. At one point in 2024, ByteDance — the parent company of TikTok — reportedly offered around $30 million to absorb the team into its own AI ecosystem. Xiao turned it down.

By April 2025, the calculus had changed dramatically. A $75 million Series B round led by U.S. venture firm Benchmark valued the company at roughly $500 million, according to CNBC. Around the same time, the company began shifting its center of gravity away from China — laying off much of its Beijing staff and relocating its headquarters and core leadership to Singapore by mid-2025, operating under a new entity called Butterfly Effect Pte. Ltd.

Eight months later, Meta announced it was buying the company outright. Estimates of the price ranged from $2 billion to as high as $3 billion, though Meta never disclosed an official figure, according to Geopolitechs. By some accounts, it ranked as Meta's third-largest acquisition ever, trailing only its purchases of WhatsApp ($19 billion) and a stake in Scale AI (roughly $15 billion), per 36Kr. Xiao Hong was set to become a Meta vice president, with his team continuing to operate as a semi-independent unit inside the company.

Why Meta Wanted It

The acquisition wasn't a side bet. It went to the heart of CEO Mark Zuckerberg's stated vision for the company's future: what he has called "personal superintelligence" — AI systems that know users deeply enough to act on their behalf, not just chat with them.

Meta's chief AI officer, Alexandr Wang — himself the founder of Scale AI before joining Meta — described Manus as providing the "execution layer" the company needed to compete with autonomous-agent projects underway at Google, Microsoft and OpenAI, according to ALM Corp. The logic was straightforward: Facebook and Instagram alone reach roughly 3.3 billion people every day, and WhatsApp adds another 2 billion. An AI agent capable of independently researching, planning, drafting and executing tasks — embedded across that user base — represented a genuinely new layer of product.

The integration moved fast. By early 2026, Meta had begun rolling Manus-derived tools into Ads Manager, the dashboard advertisers use to run campaigns across Meta's platforms, giving businesses AI-driven analysis, automated reporting and audience research, according to ALM Corp's advertiser guide. Zuckerberg, on Meta's Q1 2026 earnings call, said the company was building both a "personal agent" for everyday users and a "business agent" aimed at entrepreneurs — and that Meta AI sessions per user had already climbed by double digits following the rollout of its new in-house model family,

Muse, according to Medianama's coverage of the earnings call. Analysts also noted the deal doubled as a hedge for Meta's $160-billion-a-year advertising business at a moment when nobody is quite sure how AI agents will reshape how ads get bought, shown and paid for, per Constellation Research.

In short: the technology at the center of this dispute wasn't sitting in a lab. It was already touching the systems that serve ads to billions of people, and Meta's stated plans called for it to go much further — into "agentic shopping," automated content curation, and AI agents that act on a user's behalf across Meta's entire family of apps, according to Entrepreneur Loop.

"Singapore Washing" — and Why Beijing Said It Doesn't Work

On paper, the company Meta bought was Singaporean. Its headquarters, its registered entity, and most of its remaining leadership were in Singapore. But Manus's technology, its founders, and a large share of its institutional knowledge originated in China — and Beijing decided that mattered more than the paperwork.

China's intervention came through a mechanism known as the Foreign Investment Security Review, run jointly by the NDRC and the Ministry of Commerce — often described as Beijing's version of CFIUS, the U.S. committee that screens foreign investments for national security risk, according to Lexology and law firm O'Melveny's client alert. The review reportedly began almost as soon as the deal was announced in late December and stretched across roughly four months before the NDRC issued its order on April 27, 2026, per TheNextWeb.

Notably, O'Melveny's analysis points out that the NDRC's order referred to the acquisition of the Manus "project" — not a specific legal entity — which legal analysts read as Beijing deliberately looking past the corporate structure to "the indigenous Chinese business behind" it. The takeaway for the broader tech industry, according to the firm, is that relocating a company's headquarters and re-incorporating overseas does not place it beyond Chinese regulators' reach if its core technology, founders or research base trace back to the mainland.

The human cost of the review became visible months before the formal order. According to Bloomberg's reporting, cited by PPC Land, Chinese authorities required Manus co-founders Xiao Hong and Ji Yichao to appear before officials in Beijing in March 2026 — and both have reportedly been barred from leaving the country since.

The Unwinding

Once the order came down, Meta moved quickly to comply — at least on the surface. According to Bloomberg, as reported by the International Business Times, Meta has completed an operational separation from Manus, cut off data sharing between the two companies, and told employees to stop using Manus-built tools altogether.

An internal memo described by Bloomberg and TheNextWeb instructed staff to migrate any existing Manus-based projects onto Meta's own systems and to halt new work on the platform entirely — language the memo reportedly described as "sunsetting" Manus.

But unwinding nine months of integration is not as simple as flipping a switch. As one law firm put it, the parties are now being asked to "unscramble the eggs" — disentangling shared code, shared personnel, and, most worryingly for users, shared data that had already begun flowing between Manus's systems and Meta's internal platforms, including its advertising infrastructure, according to O'Melveny. Beijing reportedly gave the companies a deadline of "several weeks" to restore Manus's Chinese assets to their pre-acquisition state — a standard that's murky to begin with, since the company being unwound is legally Singaporean, not Chinese, per TheNextWeb.

What Happens to Manus Now

With the deal collapsing, Manus's three founders are reportedly trying to raise roughly $1 billion from outside investors to buy the company back from Meta at the same roughly $2 billion valuation Meta originally paid — with the founders potentially covering any shortfall out of their own pockets, according to Bloomberg's reporting via TheNextWeb and PPC Land. If that buyback succeeds, the plan reportedly involves reorganizing Manus as a Chinese joint venture with new backers — potentially setting the stage for a Hong Kong IPO down the road.

The financial case for the buyback rests on Manus's underlying business: the company reportedly reached more than $100 million in annualized revenue within roughly eight months of launching its general-purpose agent — a figure that helps explain why both Meta and a new round of investors see real value in the platform, independent of the geopolitical fight swirling around it, per TheNextWeb.

Why This Should Matter to Facebook Users

Here's the part that tends to get lost in the deal-making and regulatory jargon: for a period of months, an AI agent with origins in China — one capable of independently browsing, researching and executing tasks — had a foothold inside the systems that power Facebook's, Instagram's and WhatsApp's advertising infrastructure. And the U.S. national security apparatus wasn't the agency that flagged a problem with that arrangement. China's was.

That's not the only red flag Manus has drawn. Even before Meta bought it, Manus had already attracted scrutiny from privacy regulators and government agencies around the world:

• In March 2025, Tennessee became the first U.S. state to formally ban Manus from government networks and devices, lumping it together with DeepSeek as tools with "direct ties to the Chinese Communist Party." Gov. Bill Lee's office cited concerns that the platform collects sensitive user data — including IP addresses, keystrokes and behavior patterns — and stores it "on Chinese servers for unspecified purposes," warning of the potential for espionage or influence operations, according to the official announcement from the Tennessee governor's office.

• Separately, security researchers told CyberGuy that they had traced data from Manus back to servers located in Shenzhen, China — raising questions about jurisdiction and who, ultimately, can access that information.

• According to Luiza's Newsletter, Manus has also drawn attention from data protection authorities across the European Union, while officials in the U.S., Taiwan and South Korea have reportedly imposed partial restrictions on the platform over national security concerns.

None of this means the specific code running inside Meta's Ads Manager today is unsafe, or that user data has been compromised. Meta has said it has now cut Manus off from its internal systems entirely. But the episode is a reminder of something Silicon Valley has been slow to admit: when a company integrates a third-party AI system this deeply into the infrastructure serving billions of users, the question of who built it, where its roots are, and which government can compel changes to it isn't an abstract one. It's operational.

The Bigger Picture: Borders Are Back

For two decades, the dominant story about technology was that it transcended borders — that data, code, capital and talent could move freely, and that a company's legal address mattered less than its product. The Meta-Manus collapse is one of the clearest signals yet that, at least for AI, that era may be ending.

Governments on both sides of the Pacific are increasingly treating AI talent and AI systems the way earlier generations treated nuclear technology or advanced semiconductors: as strategic assets subject to national control, regardless of where the paperwork is filed. China's intervention in the Manus deal — reaching past a Singapore-incorporated company to claim jurisdiction over its "indigenous" Chinese origins — sets a template that, according to legal analysts at O'Melveny, could apply to "every subsequent case" of a Chinese-founded company trying to relocate and sell itself abroad.

The United States, for its part, has been moving in a parallel direction. An outbound investment rule that took effect in January 2025 restricts certain U.S. investments in Chinese AI, semiconductor and quantum technology firms, according to Lexology's analysis — and reporting indicates Meta structured the Manus deal in part to try to fit within an exemption to those rules by insisting on a full buyout of Manus's Chinese-origin equity. China's countermove suggests that strategy didn't fully insulate the deal from scrutiny on the other side.

The collapse also lands amid broader questions in Washington about Meta's history with China. In April 2025, former Facebook policy director Sarah Wynn-Williams testified before a Senate Judiciary subcommittee that Meta executives had, in her account, "repeatedly undermine[d] U.S. national security and betray[ed] American values" in pursuit of access to the Chinese market — allegations Meta has disputed, according to Fox Business and NBC News. That earlier controversy is a separate matter from the Manus deal, but it forms part of the backdrop against which lawmakers and analysts are now reading this latest episode.

As Winston Ma, an adjunct law professor at NYU who studies AI and the digital economy, put it in comments to CNBC, China's move amounts to "a powerful reminder" to Zuckerberg and to the broader U.S. market about how the AI race is actually going to be run — not just in research labs and product launches, but in the offices of state regulators who increasingly view AI the same way they view oil, chips or rare earths: too important to let anyone else control.

WECU Media will continue following the unwinding of the Meta-Manus deal as it develops, including what it means for the roughly 3.4 billion people who use Meta's apps every day.