China Vetoes Meta's $2bn Manus Deal: Lessons for Acquirers

What Beijing blocked and why

China's State Administration for Market Regulation (SAMR) refused to approve Meta's bid for Manus, the Shenzhen-headquartered AI agent platform, as first reported by the BBC on 27 April 2026. The deal, valued at approximately $2 billion, had been under regulatory review since late 2025.

No detailed public statement accompanied the ruling. However, people familiar with the review process told the BBC that regulators concluded the transaction would transfer nationally significant artificial-intelligence intellectual property to a foreign entity. The decision sits within an increasingly muscular Chinese regulatory framework that has expanded steadily since 2023, when Beijing tightened technology-export licensing rules and broadened its "unreliable entity list" to cover a wider set of advanced computing and AI-related assets.

The block is not without precedent. In 2018, Qualcomm's $44 billion proposed acquisition of NXP Semiconductors collapsed after SAMR withheld approval, widely interpreted at the time as a bargaining chip in the US-China trade dispute. More recently, regulators attached stringent conditions to Broadcom's $61 billion acquisition of VMware before granting conditional clearance. The Manus ruling goes further: an outright veto rather than a set of behavioural remedies.

Manus and the AI-agent land grab

Manus launched in early 2025 as a general-purpose AI agent platform, designed to let businesses deploy autonomous software agents capable of executing multi-step tasks without continuous human oversight. Founded by a team of engineers with roots in Tsinghua University's computer science faculty, the start-up attracted attention for the speed at which it moved from research prototype to commercial product.

Within months of launch, Manus had secured enterprise pilot contracts across logistics, financial services, and e-commerce sectors in China and South-East Asia, according to Chinese technology publication 36Kr. Its agent orchestration layer, which coordinates multiple large language models to handle complex workflows, was regarded by industry analysts as among the most advanced outside the major US labs.

For Meta, Manus represented a shortcut. Mark Zuckerberg, Meta's chief executive, has made AI the company's central strategic priority. Meta has invested heavily in its open-source Llama model family and, according to filings with the US Securities and Exchange Commission, spent more than $37 billion on capital expenditure in 2025, much of it directed at AI infrastructure. Yet the company has been candid about gaps in its agentic AI capabilities. Acquiring Manus would have given Meta a production-ready agent platform and, critically, a team of researchers steeped in the Chinese AI ecosystem's distinctive approach to rapid iteration and deployment.

Meta has pursued other AI acquisition targets in parallel. Reports in the Financial Times and The Information have linked the company to exploratory talks with at least two further AI start-ups in 2025 and early 2026, though none of those discussions has resulted in a confirmed deal.

Cross-border deal risk: a new normal

The Manus block crystallises a trend that dealmakers have been tracking for several years. Sovereign governments on both sides of the Pacific are treating advanced AI as a strategic asset class, subject to the same protective instincts historically reserved for defence technology, critical minerals, and telecommunications infrastructure.

China's toolkit has grown. The revised Foreign Investment Law, effective since 2020, gives regulators broad discretion to block transactions that could harm "national security or the public interest." Amendments to the Technology Export Restriction Catalogue in 2023 added several categories of AI-related IP, including recommendation algorithms, natural language processing architectures, and autonomous agent frameworks. Manus's core technology appears to fall squarely within these expanded categories.

On the other side, the United States has its own thickening web of controls. The Committee on Foreign Investment in the United States (CFIUS) has widened its scrutiny of outbound investment, and executive orders signed in 2023 and 2024 restrict US entities from investing in certain Chinese AI, semiconductor, and quantum-computing ventures. The practical result is a narrowing corridor for any cross-border AI transaction involving US or Chinese parties.

How UK and EU regulators compare

The UK's Competition and Markets Authority (CMA) has signalled its own willingness to intervene in AI-related deals. Its investigation into Microsoft's relationship with Mistral AI, opened in late 2025, examined whether a partnership structure could amount to a de facto merger. The European Commission, meanwhile, has been scrutinising large AI partnerships under its Foreign Subsidies Regulation and the Digital Markets Act, adding layers of compliance that acquirers must navigate.

Neither the CMA nor the Commission has yet blocked an AI acquisition outright. But the direction of travel is clear: regulators globally are building capacity and legal authority to intervene where AI capabilities are deemed strategically sensitive.

What UK operators should take from this

For UK founders, finance directors, and board members considering acquisitions or partnerships involving Chinese-linked AI firms, the Manus ruling offers several concrete lessons.

Map the regulatory surface early. Any deal touching Chinese AI intellectual property now faces at least three potential veto points: SAMR in China, CFIUS if a US entity is involved, and potentially the CMA or European Commission depending on where the acquirer operates or lists. Due-diligence teams should produce a regulatory-risk matrix before heads of terms are signed, not after.

Assess IP provenance with precision. Manus's technology was developed in China by Chinese nationals, placing it firmly within Beijing's export-control perimeter. Deals involving AI firms with distributed teams, split IP ownership, or dual-headquarter structures may face different, though not necessarily lighter, scrutiny. Understanding exactly where models were trained, where data resides, and where key researchers hold citizenship is now a prerequisite.

Price in delay and failure risk. The Manus review lasted several months before ending in a block. For acquirers, that represents significant management distraction and advisory cost with no completion. Deal structures should include realistic long-stop dates and reverse break fees calibrated to the genuine probability of regulatory failure, not just the theoretical possibility.

Consider alternative structures. Licensing arrangements, joint ventures, and minority stakes may attract less regulatory friction than outright acquisitions, though they are not immune. The CMA's Mistral inquiry demonstrates that regulators are prepared to look through the legal form of a transaction to its economic substance.

The Manus block does not close the door on all cross-border AI transactions. It does, however, confirm that sovereign gatekeeping of AI capabilities is now an operational reality, not a policy discussion. Boards that treat regulatory-risk assessment as a box-ticking exercise rather than a strategic discipline will find themselves on the wrong side of that reality.