China's decision to force Meta to unwind its $2 billion acquisition of Manus AI marks something new in the escalating technology war between Washington and Beijing: the formal elevation of AI data sovereignty to the same policy tier as semiconductor controls.
The ruling, announced April 27, gave Meta 90 days to divest its stake in Manus, the AI agent startup founded by Chinese entrepreneurs that used Anthropic's Claude 3.7 Sonnet as its underlying engine. Chinese regulators cited national security concerns tied to data access—but the real issue runs deeper than traditional data protection. Manus's agentic architecture enables AI systems to act autonomously on behalf of users, accumulating behavioral patterns, search preferences, and decision-making data that could prove invaluable in the global AI arms race.
The Manus acquisition closed in December 2025. Chinese regulators announced their probe in January 2026, instructing the company's two cofounders to remain in China during the investigation. The April 27 veto represents the first time Beijing has blocked a US AI acquisition using national security powers—a milestone that signals AI data has joined chips as a domain requiring state-level protection.
The precedent matters beyond this single deal. Manus's platform handles complex multi-step tasks—searching real estate listings, booking international travel, managing workflows—generating vast datasets on user behavior. Beijing has now made clear that such data, when processed by foreign-controlled AI systems, constitutes a strategic asset requiring government oversight. The question is no longer where data sits on a server, but who controls the AI models that interpret and act upon it.
For Meta, the unwinding creates immediate complications. The company had positioned Manus as a cornerstone of its AI agent strategy, leveraging the startup's agentic harness architecture to deploy Claude's capabilities across consumer applications. Losing access to Manus's team and technology leaves a gap in Meta's product roadmap that cannot easily be filled.
For Manus, the situation is precarious. The startup now faces the prospect of operating without its largest backer, potentially seeking new funding or acquisition by a domestic buyer. ByteDance, Baidu, and Alibaba—all building competitive AI agent platforms—may emerge as potential suitors for Manus's technology and talent.
The deal's collapse reflects a broader pattern. US-China technology decoupling has moved from hardware to data, algorithms, and applications. Beijing's message is unambiguous: controlling the flow of data through AI systems matters as much as controlling semiconductor supply chains. For global AI companies, this should sound a clear alarm. Cross-border AI deals will face increasing scrutiny—not just over where data is stored, but who can access the behavioral patterns and decision-making logic that AI systems generate. AI sovereignty has arrived as a concrete regulatory reality, not a theoretical concept.