On April 27, 2026, China's National Development and Reform Commission issued a one-line directive that shattered the largest cross-border AI acquisition in history. Meta Platforms, which had completed a roughly $2 billion purchase of Manus AI, was ordered to unwind the deal entirely. The target was not a Chinese company in the conventional sense -- Manus is headquartered in Singapore, operates in English, and had already been absorbed into Meta's product organization. But the NDRC claimed jurisdiction anyway, and the unwinding process began within days.
The Manus block is not an isolated regulatory action. In the same week, Beijing ordered three of its most prominent AI companies -- Moonshot AI, StepFun, and ByteDance -- to reject US capital without prior government approval. On the American side, the Treasury Department's Outbound Investment Security Program, which went into effect in January 2025, has been steadily expanding its restrictions on US investment into Chinese AI, with the COINS Act, signed in December 2025, codifying those limits into permanent law. Washington screens money going out; Beijing screens money coming in. The result is a双向夹紧 -- a two-directional squeeze that is splitting the global AI landscape into two closed, increasingly isolated ecosystems.
Who Is Manus AI
Manus -- Latin for "hand" -- is an autonomous AI agent platform developed by a company called Butterfly Effect. The platform can create slides, build websites, develop desktop applications, and execute multi-step tasks without human intervention. It is not a foundation model maker in the mold of OpenAI or Anthropic; rather, it operates as an orchestration layer that calls on existing models to complete complex workflows.
Butterfly Effect was founded in China by Xiao Hong (CEO), a graduate of Huazhong University of Science and Technology in Wuhan, along with co-founders Ji Yichao (Peak Ji), who serves as Chief Scientist, and Tao Zhang. The team built Manus in China but relocated the company's headquarters to Singapore as US chip export restrictions tightened and Chinese regulatory scrutiny of AI companies increased. This relocation pattern -- what some analysts now call "Singapore-washing" -- has become common among Chinese AI startups seeking access to global capital markets while maintaining their technical teams in mainland China.
Prior to Meta's acquisition, ByteDance had reportedly attempted to acquire Manus but was rejected. The Meta deal, valued at over $2 billion (some reports cite $2.5 billion including retention packages), was seen as a landmark validation of agentic AI -- the concept that the next wave of AI value would come not from chatbots that answer questions, but from autonomous systems that execute tasks. Reuters broke the story of China's intervention on April 27, 2026.
What Makes Manus Different from Other AI Startups
Manus occupies a distinct niche in the AI landscape. While most Chinese AI companies gaining Western attention are foundation model developers -- DeepSeek with its cost-efficient models, Moonshot AI with long-context capabilities, StepFun with multimodal systems -- Manus built its value proposition on application-level intelligence. Its agents do not train models; they use them. This distinction matters because it means the technology's strategic value lies in the orchestration logic, the task-planning architecture, and the accumulated training data from real-world task execution -- all of which Beijing apparently considers sensitive enough to block from foreign acquisition.
The relocation to Singapore added a layer of corporate complexity that ultimately failed to shield the deal. Manus maintained operations in China, employed Chinese engineers, and built its technology on the mainland. The NDRC's assertion of jurisdiction, while surprising to outside observers, follows a pattern: Chinese regulators have increasingly claimed authority over technology developed by Chinese nationals regardless of where the holding company is incorporated.
China's Block: Unscrambling the Egg
The mechanics of the Manus block are as significant as the decision itself. This was not a preemptive rejection of a proposed deal. According to multiple reports, the Meta acquisition had already been completed -- Manus had been integrated into Meta's operations. The NDRC ordered what amounts to a retroactive unwinding, forcing the separation of companies that had already been merged.
This is "unscrambling the egg" -- and it sends a chilling signal to every investor and acquirer evaluating Chinese or China-originated technology companies. The message is that regulatory approval is not a closing condition you can clear in advance; it is an ongoing, potentially retroactive authority that can be exercised at any time.
The founders of Manus reportedly face exit bans from China, preventing them from leaving the country. Exit bans are a tool Beijing has used sparingly but effectively in cases involving national security or financial investigations. Their application here signals that Chinese authorities view the Manus technology -- and the founders' knowledge -- as strategic assets that cannot be transferred to a US entity under any circumstances.
The NDRC's jurisdictional claim is worth examining. Manus is a Singapore-based company. Its corporate structure, incorporation documents, and legal domicile are all in Singapore. But the NDRC asserted authority based on the Chinese nationality of the founders, the Chinese origin of the underlying technology, and the continued presence of technical operations in China. This is an expansive interpretation of regulatory reach -- one that effectively means any technology company founded by Chinese nationals, regardless of where it relocates, remains subject to Beijing's approval for any major corporate transaction.
China's Broader Move: Three AI Companies, One Message
The Manus block did not happen in isolation. In the same regulatory sweep, Beijing ordered Moonshot AI, StepFun, and ByteDance to reject US capital without explicit government clearance. This is not a blanket ban on all foreign investment -- it is a targeted restriction on US capital specifically, applied to companies operating at the frontier of China's AI capabilities.
Moonshot AI, founded by former ByteDance vice president Yang Zhilin, has emerged as one of China's leading large language model developers, with its Kimi chatbot gaining significant domestic market share. StepFun, backed by Qihoo 360 founder Zhou Hongyi, focuses on multimodal AI systems. ByteDance, the parent company of TikTok, has been investing heavily in AI infrastructure -- as covered in our analysis of bytedance-profit-drop-ai-investment, where profits reportedly fell 70% as the company redirected resources toward AI development.
The targeting of these three companies is not random. Together, they represent the frontier of Chinese AI capability in language models, multimodal systems, and applied AI infrastructure. By restricting their access to US capital, Beijing is attempting to prevent American investors from gaining influence -- through board seats, information rights, or financial leverage -- over the direction of China's most advanced AI development.
China has also moved to ban Chinese AI companies from listing on US stock exchanges, extending what began as the Didi precedent into a comprehensive firewall. When Didi Global went public on the NYSE in June 2021 at a $73 billion valuation, Chinese regulators launched a cybersecurity review within 48 hours, forced the company to delist, and fined it $1.2 billion. The Didi case was officially framed as a data security matter, but the subtext was clear: Beijing would not allow its largest technology companies to fall under US regulatory jurisdiction or expose sensitive operational data to American oversight through the public listing process.
The US Side: OISP, CFIUS, and the "Reverse CFIUS"
While Beijing was blocking inbound investment, Washington was building its own fence. The Outbound Investment Security Program (OISP), implemented by the US Treasury Department in January 2025, established the first systematic US government review of American investment flowing into Chinese AI, semiconductor, and quantum computing companies. The program operates through two mechanisms: outright prohibitions on certain types of investment and mandatory notification requirements for others.
In practice, the OISP means that a US venture capital firm considering an investment in a Chinese AI startup must either notify the Treasury Department or, for deals involving certain sensitive technologies, is prohibited from investing at all. The program has been expanded twice -- first in December 2025 to cover additional countries of concern and critical technologies, and again through the COINS Act, which was signed into law in early 2026 and codified what had been an executive order-based program into permanent statute.
Data sources: US Treasury Dept OISP program, Reuters, Bloomberg, NDRC public notices
CFIUS (the Committee on Foreign Investment in the United States) operates on the inbound side, screening Chinese investment into US companies. The OISP is sometimes called "Reverse CFIUS" because it mirrors the same screening logic but points outward -- restricting US capital from flowing to China rather than Chinese capital from flowing to the US. Together, the two programs create a comprehensive barrier: Chinese money cannot easily invest in US AI companies, and US money cannot easily invest in Chinese AI companies.
The Treasury Department's review of the Manus deal through its Benchmark division illustrates how these mechanisms interact. Even before the NDRC intervened, the US government was examining whether Meta's acquisition of a China-originated AI company raised national security concerns. Both governments were simultaneously scrutinizing the same transaction from opposite sides -- and both reached the conclusion that the deal should not proceed.
Two Closed AI Ecosystems
The combined effect of these parallel restrictions is the emergence of two fundamentally separate AI ecosystems. In the US ecosystem, companies like OpenAI, Anthropic, Google DeepMind, and Meta itself operate with access to the world's deepest pools of venture capital, advanced semiconductor supply chains (dominated by NVIDIA), and a talent market that draws researchers from every continent. In the Chinese ecosystem, companies like DeepSeek, Moonshot AI, StepFun, Alibaba's Tongyi Qianwen, and ByteDance operate with domestic capital, increasingly capable but technically constrained semiconductor alternatives, and a talent pool that is large but increasingly isolated from international collaboration.
The capital gap between these ecosystems is stark. In 2024, US AI startups received approximately $109.1 billion in private investment, compared to roughly $9.3 billion for Chinese AI startups -- a roughly 12-to-1 disparity. By Q1 2026, the gap had widened further: US AI companies attracted $250 billion in venture funding versus $16.1 billion for their Chinese counterparts, according to data compiled by multiple research firms. Yet the Stanford AI Index 2026 found that the performance gap between top US and Chinese AI models had narrowed to just 2.7 percentage points, down from 31.6 points in 2023.
| Metric | US | China | Gap |
|---|---|---|---|
| 2024 Private AI Investment | $109.1B | $9.3B | 12:1 |
| Q1 2026 Venture Funding | $250B | $16.1B | 15.5:1 |
| AI Model Performance Gap (2023) | — | — | 31.6 pp |
| AI Model Performance Gap (March 2026) | — | — | 2.7 pp |
This paradox -- narrowing performance gap despite widening capital gap -- suggests that the current equilibrium is unstable. Either the capital advantage will eventually translate into insurmountable performance leads (as it would in most industries), or Chinese companies have discovered efficiencies that fundamentally change the economics of AI development. The answer matters enormously, and the investment restrictions make it harder for anyone outside either ecosystem to assess which scenario is playing out.
The Singapore-Washing Pattern
Manus is not the first Chinese AI company to relocate to Singapore, and it will not be the last. The pattern has become common enough to earn a name: "Singapore-washing." Chinese AI startups incorporate a Singapore holding company, move their legal domicile, and sometimes relocate key executives -- all while maintaining their engineering teams, research operations, and primary customer base in mainland China.
The motivations are practical. A Singapore headquarters provides access to international banking, makes it easier to receive USD-denominated investment, and insulates the company from some aspects of US-China regulatory friction. Singapore's legal system, intellectual property protections, and double taxation treaties with most major economies make it an attractive corporate base.
But the Manus case demonstrates the limits of this strategy. Singapore-washing can restructure a company's legal identity, but it cannot sever the regulatory claims that Beijing asserts over technology developed by Chinese nationals on Chinese soil. The NDRC's jurisdictional claim over Manus -- a Singapore-incorporated company with a Singapore headquarters -- establishes that corporate restructuring alone is insufficient to bypass Chinese regulatory oversight. Any company with substantive Chinese operations or Chinese-national founders is, in Beijing's view, subject to Chinese technology transfer controls regardless of its corporate domicile.
For investors, this creates a category of risk that is difficult to price. A Chinese AI startup incorporated in Singapore may look like a normal international investment on paper, but it carries the same regulatory uncertainty as a direct investment in a Shanghai-based company. The legal architecture provides no protection if Beijing decides to intervene.
Impact: Capital Squeeze, Deal Risk, and the Didi Precedent
The practical consequences of the Meta-Manus block and the broader investment restrictions are already visible across three dimensions.
First, Chinese AI startups face a deepening capital squeeze. Unable to list in the US, restricted from accepting American venture capital, and operating in a domestic funding environment that is a fraction of the size of Silicon Valley's, these companies must rely on Chinese state-linked funds, domestic tech giants like Alibaba and Tencent, and a limited pool of international investors willing to navigate the regulatory complexity. The 12-to-1 capital gap cited earlier is not just a statistical abstraction -- it translates into fewer GPUs, smaller research teams, and slower iteration cycles.
Second, deal certainty has collapsed for any transaction involving Chinese AI assets. The Manus unwinding demonstrates that even completed deals are not final. The NDRC's retroactive intervention -- ordering the separation of two companies that had already merged -- creates a category of risk that standard M&A legal frameworks were not designed to handle. Acquirers must now price in the possibility that Chinese regulators will unwind a deal after closing, after integration has begun, and after confidential technology and personnel information has been transferred.
Third, the Didi precedent has been extended from public markets to private transactions. When Beijing forced Didi to delist from the NYSE in 2021, it established that Chinese regulators would exercise authority over companies' access to US capital markets. The Manus block extends this principle to private M&A, signaling that any transfer of Chinese-developed AI technology to a US entity -- whether through acquisition, investment, or licensing -- requires Beijing's explicit approval.
Photo credit: Luca Bravo on Unsplash
What This Means for Global AI Development
The fragmentation of the global AI ecosystem has implications beyond investment returns. Knowledge transfer between the two AI ecosystems has slowed dramatically. Chinese researchers publish fewer papers in Western venues, and American researchers cite Chinese work less frequently -- not because the work is inferior, but because the institutional channels that previously facilitated exchange have been systematically dismantled.
Talent flows have also shifted. Chinese AI researchers who previously split their careers between US and Chinese institutions increasingly face binary choices: work in the US ecosystem with access to cutting-edge hardware and massive capital, or work in the Chinese ecosystem with access to vast domestic datasets and lower regulatory barriers to deployment. The number of researchers who can operate in both ecosystems is shrinking, and with it, the cross-pollination of ideas that historically accelerated progress in both.
The speed of development in each ecosystem is also diverging in character. The US ecosystem, with its capital abundance, tends to pursue scale -- larger models, more compute, bigger training runs. The Chinese ecosystem, constrained by both capital and chip access, has been forced to pursue efficiency -- smaller models that achieve comparable results with fewer resources. As covered in our analysis of china-ai-token-usage-scale, China's daily AI token consumption has surpassed 140 trillion, a 1,000-fold increase in two years, driven largely by cost-efficient models deployed at massive domestic scale.
Both approaches produce advances, but they are increasingly advances that the other side cannot easily replicate or build upon. The US cannot easily match Chinese deployment speed and cost efficiency without the domestic demand and regulatory permissiveness that drives it. China cannot easily match US frontier model capabilities without the GPU clusters and capital that underpin them. The two ecosystems are not just separated -- they are evolving along different optimization paths.
By China Made & Tech Team. Independent publication covering Chinese manufacturing and technology innovation for global audiences.
Methodology note: This analysis draws on reporting from Reuters, Bloomberg, the Wall Street Journal, and official US Treasury Department publications regarding the OISP program. Capital comparison figures are derived from Stanford AI Index 2026 data and industry research from multiple venture capital tracking firms. The 12:1 capital ratio refers to 2024 private AI investment figures; Q1 2026 data shows a widening gap.
Frequently Asked Questions
Why did China block Meta's acquisition of Manus AI?
China's NDRC blocked the deal asserting jurisdiction over technology developed by Chinese nationals, even though Manus was incorporated in Singapore. Beijing views advanced AI agent technology as a strategic asset and invoked regulatory authority based on the founders' Chinese nationality and the technology's Chinese development origins, ordering Meta to unwind the completed $2 billion acquisition.
What is the Outbound Investment Security Program (OISP)?
The OISP is a US Treasury Department program effective since January 2025 that restricts American investment into Chinese AI, semiconductor, and quantum computing companies. It operates through outright prohibitions on certain investments and mandatory notification requirements for others. The COINS Act, signed into law in December 2025, codified the program into permanent law.
What does "Singapore-washing" mean for AI companies?
Singapore-washing refers to Chinese AI startups relocating their legal headquarters to Singapore to access international capital and reduce regulatory friction, while maintaining engineering teams and operations in mainland China. The Manus case demonstrates this strategy's limits: Beijing claimed regulatory jurisdiction over the Singapore-incorporated company based on the Chinese nationality of its founders and the technology's mainland origins.
How big is the US-China AI investment gap?
In 2024, US AI startups received approximately $109.1 billion in private investment versus roughly $9.3 billion for Chinese AI startups -- a 12-to-1 ratio. By Q1 2026, the gap widened to approximately $250 billion versus $16.1 billion, a roughly 15.5-to-1 disparity. Despite this, the AI model performance gap between the two countries narrowed to just 2.7 percentage points.
Which Chinese AI companies were ordered to reject US investment?
Beijing specifically ordered Moonshot AI, StepFun, and ByteDance to reject US capital without prior government approval. These three companies represent the frontier of Chinese AI capability in language models, multimodal systems, and applied AI infrastructure respectively.
Related Entries
- china-ai-robotics-guide -- China AI industry overview
- china-ai-regulation -- AI regulation landscape
- deepseek-v4-launch-analysis -- DeepSeek's latest model launch
- bytedance-profit-drop-ai-investment -- ByteDance's AI spending and profit decline
- china-ai-token-usage-scale -- China's AI infrastructure scale