China kills Meta’s Manus deal — $2B down the drain

China just pulled the plug on Meta’s biggest AI bet.

The country’s antitrust regulator ordered Meta to unwind its $2 billion acquisition of Manus, an AI agent startup that Zuckerberg had been chasing for months. The deal was supposed to supercharge Meta’s push into autonomous AI agents — the kind that can book flights, send emails, and manage your calendar without you lifting a finger.

Instead, Meta is walking away empty-handed. And Zuckerberg’s AI agent strategy just took a serious hit.

What happened

The probe started back in late 2025, not long after the deal was announced. Chinese regulators had concerns about data security, market concentration, and whether Meta would have too much control over AI agent technology that could be used in sensitive sectors.

I’ve seen this pattern before. China has been tightening its grip on AI-related acquisitions for years, especially when foreign companies are involved. But blocking a deal this size — $2 billion is no small sum — signals that Beijing is taking AI agent technology very seriously.

The official statement cited “national security risks” and “potential harm to fair competition.” Vague, but that’s how these things usually go. The real issue is probably that Manus had access to massive amounts of Chinese user data and its AI models were trained on local datasets. China doesn’t want that flowing into Meta’s hands.

Why this hurts Meta

Meta has been playing catch-up in AI. While OpenAI, Google, and even Microsoft have been shipping AI features for years, Meta’s efforts have been scattered. They have the open-source Llama models, but that’s more of a research play than a product.

Manus was supposed to change that. The startup had built a genuinely impressive AI agent platform — think of it as a souped-up version of what Anthropic’s Claude can do, but focused on enterprise workflows. Meta wanted to integrate that into WhatsApp, Messenger, and maybe even Facebook itself.

Without Manus, Meta’s AI agent roadmap looks thin. They could build their own, but that takes years. Or they could try to acquire another startup, but the pickings are slim. Most of the good ones are already snapped up or based in countries that won’t sell to US tech giants.

The bigger picture

This isn’t just about Meta. It’s a signal that China is drawing a line in the sand on AI technology transfers. If you’re a US company eyeing Chinese AI startups, good luck. The regulatory environment is hostile, and even if you get a deal signed, there’s no guarantee it will close.

Meanwhile, Chinese AI companies like Baidu, ByteDance, and a bunch of well-funded startups are building their own agent ecosystems. They don’t need Meta’s money or expertise. In fact, they’re probably relieved the deal fell through — less competition.

I find this ironic. Meta spent years trying to win over China, from Zuckerberg’s Mandarin speeches to the ill-fated Facebook Watch launch. None of it worked. And now, even a $2 billion acquisition can’t buy goodwill.

What’s next

Meta has 90 days to unwind the deal. That means returning any assets, data, or intellectual property it already transferred. It’s messy, expensive, and embarrassing.

Zuckerberg will probably pivot to Plan B — whatever that is. Maybe they’ll double down on their own AI agent research, or look for targets outside China. But the clock is ticking. The AI agent race is heating up, and Meta just lost a major shortcut.

Personally, I think this might actually be a blessing in disguise for Meta. Manus was expensive, and integrating a Chinese startup into Meta’s infrastructure would have been a regulatory nightmare. But that’s a cold comfort when your competitors are shipping products and you’re stuck rebuilding from scratch.