When AI startups leave home: Manus, Meta and the reality of cross‑border AI deals

An elegant demo, a rapid funding round and a headline exit—classic startup theater. Manus followed that arc: an early showcase of AI agents and AI automation for business, a $75 million round led by Benchmark (reported), millions of users and more than $100 million in annual recurring revenue (reported), and then a roughly $2 billion acquisition by Meta as part of the company’s wider push on AI. The twist came after the curtain call: geopolitics pulled the stage ropes.

Quick definitions

• AI agents: software that autonomously carries out tasks—scheduling, recruiting, customer outreach—by combining models, tooling and automation.
• AI automation: applying AI to replace or augment repeatable business workflows (e.g., sales outreach, HR screening).
• ARR (annual recurring revenue): subscription revenue run‑rate used to value SaaS and AI businesses.
• Foreign‑investment review: a government process that examines cross‑border deals for national security or strategic risks (China’s NDRC and the U.S. CFIUS are two examples).

Timeline (high‑level)

• Demo: Manus surfaces with a demo of an AI agent handling recruiting, travel planning and portfolio analysis.
• Funding: Benchmark reportedly leads a $75M round valuing Manus at ~ $500M.
• Growth: Manus claims millions of users and >$100M ARR (reported).
• Relocation: Manus moves its core team and legal domicile from Beijing to Singapore and restructures ownership.
• Acquisition: Meta acquires Manus for ~ $2B; Meta pledges to cut ties with Chinese investors and close Manus’s China operations.
• Regulatory review: Chinese authorities summon Manus founders and the NDRC reportedly opens an inquiry; founders were reportedly told they could not leave the country during the review.

What happened and why it matters

Manus was not just another app or tool. It offered an early glimpse of AI agents that could automate parts of hiring, travel and even financial analysis—features that matter to enterprise buyers looking to scale productivity and to buyers hunting differentiated AI capabilities. That made Manus strategically attractive to a company like Meta, which is aggressively building AI for business and consumer platforms.

But strategic value breeds strategic concern. China has been tightening oversight of technologies it considers nationally important. Regulators framed the scrutiny as routine, but their concern is clear: rapid outflows of talent, IP and strategic capabilities—what some critics call “selling young crops” (selling promising domestic startups to foreign buyers)—could undermine domestic technological development and national security.

“Financial Times reported that Manus’s founders were told they would be prevented from leaving the country while authorities examine the Meta transaction.”

On the other side of the Pacific, U.S. lawmakers have warned that investment in certain Chinese AI capabilities can strengthen a strategic rival. Senator John Cornyn urged caution, arguing that some investments may “effectively subsidize a geopolitical rival” (reported). The Manus episode thus sits at the intersection of business strategy and geopolitics: cross‑border AI M&A is no longer purely commercial.

What is a foreign‑investment review?

• Who conducts it: Typically a national body (e.g., China’s National Development and Reform Commission, the U.S. Committee on Foreign Investment in the United States).
• Why it matters: Reviews assess whether deals transfer sensitive technology, critical infrastructure, or talent in ways that harm national security or economic interests.
• Possible outcomes: approvals, approvals with conditions (carve‑outs, restrictions), delays, or outright blockages. They can also be used as leverage to extract concessions.

Practical implications for executives and investors

The Manus episode should change assumptions for founders, acquirers and in‑house procurement teams evaluating AI vendors. Legal form and press releases don’t immunize a deal from political scrutiny. Here’s what matters now, and what to do about it.

Immediate checklist (first 30–90 days)

• Model provenance audit: Where were models trained? What datasets were used? Any restricted or government‑sourced data?
• IP chain-of-title: Ensure founders actually own the model weights and code; check for encumbrances, prior grants or research partnerships.
• Employee and contractor locational risk: Map where engineers and key contributors are legally based—country of work can create jurisdictional claims.
• Investor origins and capital flows: Identify investors from countries of concern; their involvement can trigger reviews.
• Compute and data residency: Locate where training and production compute runs; cloud regions can create exposure to export controls and data laws.

Mitigations and transaction design

• Structural: Consider licensing model weights instead of an outright transfer, staged ownership changes, or escrow arrangements for sensitive assets.
• Operational: Host critical models in neutral or local cloud regions, ringfence sensitive teams, and implement dual‑control access to key assets.
• Legal & regulatory: File voluntary notifications with relevant authorities, negotiate regulatory covenants into purchase agreements, and prepare contingency carve‑outs for restricted jurisdictions.

Business scenarios and likely outcomes

• Delay and conditional approval: Most common. Regulators slow the deal and require concessions (data localization, divestitures, or governance changes).
• Blocking or reversal: Less common but possible for tech seen as directly enabling military or critical infrastructure capabilities.
• Negotiated split: Buyer retains global rights except for a blocked jurisdiction, or forms a joint venture with local partners to continue limited operations.

Balanced view: not every deal will fail

It’s important to avoid alarmism. Governments still approve many cross‑border technology deals, often with conditions. Regulatory scrutiny doesn’t mean the end of global cooperation—rather, it raises the bar for transparency and compliance. For buyers willing to design transactions with regulatory realities in mind, cross‑border AI M&A remains viable.

Key questions for leaders

Will Beijing block or reverse the Meta acquisition?

They can delay, condition or complicate the deal via a foreign‑investment review; an outright reversal is rarer but possible. Outcomes often depend on political judgment as much as legal criteria, so prepare for contingencies and concessions.

How will this affect future cross‑border AI investments and exits?

Expect longer timelines, more scrutiny and a higher transactional cost. Some acquirers may prioritize domestic targets or structure deals to minimize perceived strategic risk.

Can startups legally insulate themselves by relocating?

Changing legal headquarters and re‑registering in another jurisdiction are useful tactics but not foolproof. Home‑country regulators can still assert jurisdiction over talent, IP and capital flows.

What does this mean for talent mobility in AI?

Travel and exit frictions will increase; founders and researchers may face constraints, and firms may favor hiring within aligned jurisdictions to reduce exposure.

Will other countries adopt similar controls?

Yes. As AI is seen as a strategic asset, more nations will tighten foreign‑investment reviews and export controls, reshaping where models are developed and hosted.

Actionable next steps

1. Run a cross‑functional AI M&A audit: legal, IP, security, HR and data teams should jointly map origin, ownership and risk.
2. Build regulatory playbooks: standard clauses, escrow terms and staged transfers that can be deployed when deals involve sensitive assets.
3. Invest in provenance tooling: track model training lineage, datasets and contributor contracts to speed diligence and demonstrate governance.
4. Engage early with counsel and regulators: voluntary disclosures can shorten timelines and reduce the risk of surprises.
5. Consider partnership models: licensing, joint ventures or local partnerships can preserve value while lowering political friction.

Manus showed how powerful AI agents can be—and how quickly value can accrue. It also reminded the market that code and talent sit inside geopolitical gravity wells. For executives building or buying AI for business, the question is less about whether automation will transform operations and more about where those systems live, who controls them, and how to design deals that survive political scrutiny. Start by treating national interest as a due‑diligence line item; it will save time, money and reputational risk down the road.

If you’re leading an AI acquisition or assessing a foreign AI vendor, run a 10‑point regulatory and provenance checklist immediately and consult counsel experienced in cross‑border AI deals. That upfront work separates headline risk from strategic opportunity.