Xiao Hong's Manus is in talks that could make Tencent its largest shareholder, Reuters reported, citing the Financial Times, after Beijing ordered Meta to unwind its roughly $2 billion acquisition of the AI agent maker.
The reported talks, published by Reuters on July 9th and carried in a URL dated July 10th, have not produced a disclosed stake size, price, valuation or signed agreement. That missing price matters. Manus was the rare Chinese-origin AI application company to command a multibillion-dollar U.S. Big Tech exit. If Tencent becomes the controlling outside backer, the practical outcome would be a reset of Manus' cap table around a Chinese technology champion that had already been reported as an earlier backer of Butterfly Effect, the company behind Manus.
For Xiao, who is publicly known as Red Xiao, the shift is a forced rerouting of a founder story that had moved fast enough to attract global capital, geopolitical scrutiny and a blocked exit in roughly 16 months. Manus launched in invitation preview in March 2025 as a general-purpose AI agent, a product category pitched around software that can carry out multi-step tasks rather than answer prompts in a chat window. By December 2025, Meta had reportedly agreed to buy Manus for about $2 billion. By April 2026, Chinese authorities had demanded that the deal be unwound.
Xiao built Manus as an application bet
Xiao's founding thesis was unusually clear for the Chinese AI wave that followed DeepSeek: Butterfly Effect would build applications on top of existing models, rather than compete in the capital-intensive race to train foundation models. In a March 2025 South China Morning Post interview, Xiao said he had not considered building large language models "from day one" because he wanted to focus on applications. SCMP also reported that Manus was built on existing models including Anthropic's Claude and fine-tuned versions of Alibaba's Qwen, citing co-founder and chief scientist Ji Yichao.
That choice made Manus easier to scale as a product and harder to classify cleanly in a cross-border acquisition fight. Manus' value sits in workflow design, orchestration, data handling and an execution environment that can run tasks in virtual computers. The core technology may rely on models from other companies, yet the application layer is exactly where users, usage data and agent behavior accumulate. Beijing's move against the Meta deal shows regulators treating that layer as strategic technology.
The founder background helps explain why Tencent is a logical buyer of influence. Xiao had already built for the WeChat economy before Butterfly Effect. Public reporting has tied his earlier Wuhan Nightingale Technology work to WeChat ecosystem tools such as Yiban Assistant and Weiban Assistant, with Tencent and ZhenFund among reported backers. Manus is a different company at a different scale, but Tencent's interest follows a pattern: back the founder early, then regain position when a foreign exit runs into state pressure.
The numbers remain mostly company-supplied
Manus has claimed the kind of traction that makes the ownership fight commercially rational. In a company blog post, Manus said it crossed $100 million in annual recurring revenue eight months after launch, had a total revenue run rate above $125 million, was growing more than 20% month over month after its Manus 1.5 release, had processed more than 147 trillion tokens and had created more than 80 million virtual computers. Manus also said it had 105 employees across Singapore, Tokyo and San Francisco, with Paris planned.
Those figures are useful, but they are Manus' figures. The blog post also said Benchmark led a $75 million round before launch and that Benchmark general partner Chetan Puttagunta joined the board. TechCrunch reported in April 2025 that Benchmark's round valued Manus at about $500 million. The reported Meta acquisition price of roughly $2 billion would have represented a steep markup within months, before Beijing forced the transaction into reverse.
The unwind has been messy because the Meta transaction appears to have moved beyond announcement. TechCrunch reported on June 13th that Meta had begun dismantling the acquisition by separating operations and halting data sharing. The Business Times reported in May that Manus' three founders, Xiao Hong, Ji Yichao and Zhang Tao, had explored raising about $1 billion from outside investors to buy back the operation.
A Tencent-led outcome would simplify one problem and create another. It could give Beijing a domestically acceptable anchor shareholder without requiring Manus' founders to finance the entire reversal themselves. It would also test whether Manus can keep selling a globally oriented AI agent while its controlling economics move closer to a Chinese internet conglomerate already under the same political system that blocked the Meta sale.
Manus is competing in agents while fighting over control
The ownership fight lands as AI agents have moved from demos into the product roadmaps of every major AI platform. Manus' early claim was that it could execute tasks such as research, website creation, itinerary planning and analysis through a cloud-based working environment. The pitch resonated because many enterprise users had already reached the limit of chatbots that draft text and require humans to finish the work.
The competitive problem is that the largest model companies are now absorbing the agent interface into their own products. OpenAI, Anthropic, Google, Perplexity and a long list of venture-backed agent companies are all chasing the same operator-style workflows: browse, click, read files, write code, assemble reports and hand back finished work. Manus' application-first approach gave Xiao speed. It also left Manus exposed to platform competition from the model providers underneath it.
That is why the Tencent talks are more than cleanup from a blocked deal. Tencent would be buying influence in one of the few Chinese-origin AI agent companies that showed fast global traction before the giants fully converged on the category. Manus would get a shareholder with distribution, capital and political acceptability in China. Meta would lose a direct route to an agent product it was willing to buy at a reported $2 billion price.
For founders building AI companies across jurisdictions, Manus is now the case study. Incorporation, staff relocation and foreign acquirers may reduce some business friction. They do not erase where a technology was built, who trained the team, who funded the first checks or which government claims the asset as strategically important. Xiao built Manus as an application company, avoiding the model race. The Tencent talks show that in AI, even the application layer can become a sovereignty fight.