China's Regulatory Authority Enters Dispute with Meta Over Artificial Intelligence Operations

On April 27, 2026, China's National Development and Reform Commission issued an order that reverberated through the global AI industry: Meta Platforms must unwind its roughly $2 billion acquisition of Manus, a Singaporean AI startup with deep Chinese roots [1]. The decision, delivered through the NDRC's Office of the Working Mechanism for Foreign Investment Security Review, represents the most consequential use of China's investment review powers in the artificial intelligence sector to date [2].

The move has frozen a deal that was supposed to accelerate Meta's push into agentic AI — systems that can autonomously execute complex tasks without step-by-step human direction. Instead, it has become a flashpoint in the intensifying contest between the United States and China over who controls the people, code, and data behind the next generation of AI.

The Deal That Triggered a Crisis

Manus burst onto the AI scene on March 6, 2025, when its invitation-only beta launch drew over a million views within twenty hours [3]. The platform, developed by Beijing-based Butterfly Effect Pte Ltd, offered something distinct from typical chatbots: an autonomous agent capable of browsing the web, writing and executing code, performing market research, and delivering completed work products without continuous human supervision [3].

Founded by Xiao Hong (CEO) and Ji Yichao (chief scientist), Manus grew rapidly, raising $75 million in a round led by Benchmark and reaching a $500 million valuation [4]. By the time it hit $100 million in annual recurring revenue, Meta came calling [4].

But the acquisition's path to completion was far from straightforward. In the months before Meta's December 2025 announcement, Manus executed a rapid corporate restructuring: it relocated its headquarters and senior engineering team from Beijing to Singapore, re-domiciled its intellectual property, and reconstituted itself as a Singaporean entity with no Chinese ownership [5]. Meta announced the $2 billion deal in December 2025, stating Manus would shut down its remaining Chinese operations [1].

Chinese authorities were not persuaded by the corporate shell game. In January 2026, China's Ministry of Commerce opened a formal review of the acquisition, examining compliance with export controls, technology import-and-export regulations, and overseas investment rules [1]. By March, the NDRC had summoned both Xiao and Ji to Beijing, where they were questioned about potential violations — and then informed they could not leave the country [6].

The Mao-Era Regulator With Modern Ambitions

The institution at the center of this dispute has roots stretching back to the earliest years of the People's Republic. The NDRC traces its lineage to the State Planning Commission, established in 1952 as a Soviet-style central planning body during the Mao Zedong era [7]. Over the decades, it evolved through multiple reorganizations — losing influence as China embraced market reforms, then gaining new powers as the government sought to manage the economy's intersection with national security.

Today's NDRC retains its mandate as China's most powerful macroeconomic planning body, but the Manus case reveals a significant expansion of its reach. The foreign investment security review mechanism it administers — sometimes described as China's equivalent of CFIUS, the Committee on Foreign Investment in the United States — has traditionally focused on sectors like defense, energy, and critical infrastructure [8]. The Manus decision signals that AI, talent, and data now fall squarely within its definition of national security [8].

The NDRC's cancellation order is notable for what it targets. Manus was, by the time of the acquisition, legally a Singaporean company. Beijing's position, articulated through state-backed media, is that corporate domicile is irrelevant — what matters is "the extent of its connections to China in terms of technology, talent, and data," and whether the transaction threatens China's "industrial security and development interests" [9].

Meta's Financial Exposure to China

Meta's relationship with China is paradoxical. Facebook, Instagram, and WhatsApp have been blocked in mainland China since 2009. Yet China has become one of Meta's most important advertising markets.

Source: Reuters / Meta SEC Filings

Data as of Apr 28, 2026CSV

Internal documents reviewed by Reuters revealed that Meta earned more than $18 billion in advertising revenue from China-based advertisers in 2024, representing over 10% of the company's total global revenue [10]. Chinese e-commerce companies like Temu, Shein, and others have spent heavily on Meta's platforms to reach consumers in the United States, Europe, and Southeast Asia. Meta's 2025 annual report acknowledged the company generates "meaningful revenue from a small number of resellers serving advertisers based in China" and flagged regulatory action as a material risk [11].

The revenue picture is complicated by quality concerns. Meta's own internal audits found that approximately 19% of its China-sourced ad revenue in the first half of 2024 — more than $3 billion — came from ads promoting scams, illegal gambling, and other prohibited content [10]. The company cut that figure roughly in half before it climbed back to about 16% by mid-2025 [10].

Analysts at several firms estimated that U.S.-China tariffs imposed during 2025 could reduce Meta's advertising revenue by up to $7 billion, as Chinese retailers pull back on overseas ad spending [12]. The Manus dispute adds another layer of uncertainty to a financial relationship already under strain.

The Legal Architecture Behind the Block

China's authority to intervene rests on an expanding body of law. The foreign investment security review mechanism, which came into effect in 2021, gives the NDRC and the Ministry of Commerce joint oversight of deals that touch national security [8]. But the Manus case also implicates newer legal frameworks.

China's 2023 Interim Measures for the Management of Generative Artificial Intelligence Services — commonly called the Gen AI Measures — assert jurisdiction over any generative AI service offered to the Chinese public, regardless of where the provider is based [13]. The 2022 Algorithm Recommendation Regulations require companies to register their algorithms and submit to government review. And amendments to China's foundational Cybersecurity Law, which took effect on January 1, 2026, significantly broadened extraterritorial enforcement, covering any overseas organization or individual whose activities harm Chinese cybersecurity [14].

The amended Cybersecurity Law also marks the first time AI governance has been formally written into one of China's foundational cybersecurity statutes, elevating AI from a regulatory concern to a legislative priority [14]. Penalties for violations can now reach RMB 10 million (approximately $1.4 million) for businesses and RMB 1 million (approximately $140,000) for individuals [14].

China's Domestic AI Compliance Regime

To evaluate whether Meta faces a double standard, it is useful to examine how China regulates its own AI companies. The Cyberspace Administration of China (CAC) requires domestic firms to submit security assessments and receive government approval before releasing mass-market generative AI products — a requirement with no parallel in the United States or European Union [15].

Source: Cyberspace Administration of China / DigiChina

Data as of Jan 15, 2026CSV

Baidu, Alibaba, and ByteDance were among the first companies to receive approvals for their generative AI products in August 2023 [16]. By 2025, China had approved over 40 large language models for public use [17], and cumulative algorithm registrations under the 2022 regulations had grown from approximately 30 in 2022 to an estimated 1,100 by 2025 [18]. Companies must file detailed information about their service forms, application fields, algorithm types, and training data sources [15].

The compliance burden is not trivial. ByteDance, which operates TikTok internationally and Douyin domestically, faces particularly complex requirements, as China's AI framework demands restructuring of how its AI models interact with content curation [15]. The system's defenders argue that Meta would face the same rules if it operated AI services in China — the company simply never has, because its core platforms have been blocked for nearly two decades.

Critics counter that the Manus case is qualitatively different. Beijing is not regulating an AI service offered within its borders; it is asserting control over a company that deliberately moved itself and its technology out of China. The dispute is less about compliance with AI regulations and more about Beijing's claim that Chinese-origin technology and talent remain within its jurisdiction regardless of where they relocate.

The Precedent Problem: From Didi to Manus

China's regulatory interventions against technology companies have escalated in scope and ambition since 2020. Alibaba was fined a record $2.8 billion in April 2021 for antitrust violations [19]. Two months later, the CAC suspended ride-hailing giant Didi's app just two days after its New York Stock Exchange IPO, citing data security violations; Didi was eventually fined over $1 billion [20]. Thirty-four technology firms, including Tencent, Meituan, and Baidu, were summoned for regulatory "rectification" sessions in April 2021 [19].

But those cases targeted companies operating within Chinese territory. The Manus case extends the logic further. Chinese officials reviewing the acquisition described it as a "conspiratorial" attempt to "hollow out the country's technology base" [1]. The state-backed Global Times framed the issue not as a question of Manus's incorporation or management location but as a matter of protecting China's engineering talent pipeline and technology stack [9].

Legal analysts have drawn parallels to CFIUS reviews in the United States, but with a critical difference: CFIUS blocks foreign acquisitions of American companies, while the NDRC is blocking a foreign acquisition of a company that had already left China [8]. This assertion of jurisdiction over departed companies and their founders — including the travel ban on Xiao and Ji — represents a significant expansion of China's regulatory reach.

What Leverage Does Each Side Hold?

The enforceability question cuts both ways. Meta has no legal presence in mainland China, which means the NDRC cannot seize assets or impose fines in the conventional sense. However, Meta does have meaningful vulnerability. Its $18 billion annual advertising revenue from Chinese clients could be disrupted if Beijing pressured Chinese companies to redirect their ad spending [10]. Meta's hardware supply chains also pass through Chinese manufacturing — though the company has been diversifying its supply base.

China's leverage over the Manus founders is more direct. Xiao Hong and Ji Yichao remain in China under a travel ban [6]. If Beijing treats the acquisition as a completed violation of its foreign investment laws, the founders could face criminal liability. The NDRC's order demands a full unwind, but the mechanics are unclear: Manus's value lies in code, models, agent workflows, and engineering know-how — assets that are difficult to return once transferred [5].

Meta, for its part, has signaled it intends to comply with the order through negotiation rather than confrontation, reflecting the company's broader interest in maintaining its Chinese advertising revenue [11].

The Wider Chill: What This Means for Cross-Border AI Deals

The Manus case has already sent shockwaves through the cross-border deal market. Reuters analysis found that the decision "raises risk for cross-border China tech deals" broadly, as investors and founders recalibrate the feasibility of relocating Chinese-origin AI companies to friendlier jurisdictions [21].

The implications extend beyond Meta. OpenAI, Google DeepMind, and Anthropic face their own China-related risks, though from a different angle. In April 2026, the three companies began collaborating through the Frontier Model Forum to combat "adversarial distillation" — attempts by Chinese AI labs to extract capabilities from leading U.S. models through unauthorized API access [22]. Anthropic documented 16 million unauthorized exchanges from DeepSeek, Moonshot AI, and MiniMax via approximately 24,000 fake accounts [22]. U.S. officials estimate this model-copying costs American AI labs billions annually [22].

None of these companies have reported receiving direct regulatory inquiries from Chinese authorities in the manner Meta has faced. But the legal exposure is real: China's amended Cybersecurity Law and Gen AI Measures assert jurisdiction over any entity whose AI services affect the Chinese public or whose activities harm Chinese cybersecurity, regardless of where the entity is based [14][13]. If China were to invoke these provisions against a U.S. AI company with no Chinese operations, it would represent a further escalation of extraterritorial regulatory claims.

The Sovereign Interest Argument

There is a case — and it deserves engagement rather than dismissal — that China's actions reflect legitimate sovereign concerns. AI talent is a genuinely finite resource. If a pattern emerges in which Chinese-trained engineers found companies in China, develop core technology using Chinese infrastructure and data, and then relocate to Singapore or the United States for acquisition by American tech giants, Beijing loses both the human capital and the technology without receiving any of the economic benefits.

From this perspective, the Manus restructuring looks less like ordinary corporate relocation and more like regulatory arbitrage: a deliberate effort to move assets beyond Beijing's reach while retaining the value created within China's technology ecosystem. Any government — including the United States through CFIUS — asserts the right to review transactions that could drain nationally significant technology capabilities.

The counterargument is equally forceful: if Chinese founders cannot freely leave the country with their own ideas and skills, the result is a system that treats human capital as state property. Travel bans on entrepreneurs who relocated their company legally undermine the rules-based order that enables global technology collaboration. And using investment review powers retroactively — blocking a deal months after it closed — introduces uncertainty that will discourage foreign investment in any company with Chinese founders or employees.

What Happens Next

The immediate question is whether Meta will comply with the unwind order and, if so, what "unwinding" an AI acquisition actually means in practice. Manus's technology has presumably been integrated into Meta's systems over the past four months. Separating code, models, and workflows is not like returning a factory.

The broader question is structural. The Manus case establishes a precedent that Chinese-origin technology companies remain subject to Beijing's jurisdiction even after relocating abroad. If that principle holds, it will reshape how venture capital flows to Chinese-founded startups, how AI talent moves between countries, and how acquirers evaluate regulatory risk.

For Meta, the immediate financial exposure is manageable — $2 billion is significant but not existential for a company that generated over $160 billion in revenue in 2025. The strategic cost is harder to measure. Manus represented a shortcut to competitive agentic AI capabilities that Meta cannot easily replicate internally.

For China, the Manus decision is a statement: in the global AI race, Beijing will treat its technology base, its trained engineers, and their innovations as national assets — and it will use the full weight of its regulatory apparatus, from Mao-era planning institutions to 2026 cybersecurity amendments, to keep them from leaving.