In December 2025, Meta paid $2 billion to acquire a Chinese AI company that most Westerners had never heard of. By April 2026, Chinese regulators had blocked the deal and forced Meta to unwind the acquisition. Now, Chinese investors led by Tencent are buying back those same shares for the original $2 billion price—except the company's revenue has grown 4-5x during the brief period under Meta's ownership.
This is the story of Manus, and it represents one of the most fascinating chapters in the ongoing competition between the United States and China for dominance in artificial intelligence. More than just a corporate drama, Manus illustrates a fundamentally different approach to AI—one that prioritizes execution over conversation, and product capability over raw model power.
What Is Manus? Understanding the "Agent Everything" Vision
Manus is not another chatbot. While most AI assistants respond to questions or generate text, Manus takes actions. It can autonomously browse the web, fill out forms, book travel, conduct market research, analyze data, and execute complex multi-step tasks—all without human intervention after receiving initial instructions.
The company's tagline is "Agent Everything," reflecting its ambition to bring autonomous AI agents to every workflow and every industry. Unlike traditional AI that requires humans to interpret results and take next steps, Manus is designed to complete entire processes from start to finish.
"Manus doesn't compete with large language models—it builds on top of them. Its core strength isn't a foundation model, but a thick Agent architecture and execution environment." — Industry analyst review
The Acquisition Drama: Meta's $2 Billion Bet
In late 2025, Meta announced the acquisition of Manus for approximately $2 billion, making it Mark Zuckerberg's third-largest acquisition behind WhatsApp and Scale AI. At the time, Manus had achieved annualized recurring revenue of just $100 million but demonstrated remarkable growth potential in the emerging AI Agent market.
The strategic rationale was clear: Meta wanted to own the "Agentic AI" category, and Manus represented the most commercially successful general-purpose AI Agent on the market. During Meta's brief ownership (roughly 4-5 months), the company integrated some of Manus's technology and helped grow its revenue to $400-500 million annualized.
However, in April 2026, China's National Development and Reform Commission's foreign investment security review office issued a rare veto, ordering both parties to unwind the deal completely. This was the first publicly blocked AI-related foreign acquisition since China's foreign investment security review mechanism was established.
The Buyback: Tencent Leads Chinese Capital's Return
The transaction's unwinding created a complex situation: Meta held all the shares, but early investors had already distributed the acquisition proceeds to their limited partners and couldn't simply return the money. Chinese investors devised a creative solution—they would simply buy back the shares at the original price.
Tencent, Sequoia China, and ZhenFund (Manus's original investors) formed a consortium to repurchase Meta's stake. The deal value remains locked at $2 billion—no premium, no discount—despite Manus now generating 4-5x more revenue than when Meta first acquired it. Financial observers describe this as "buying a high-growth asset at a significant discount."
American venture capital firm Benchmark, which had participated in Manus's earlier funding rounds, chose not to participate in the buyback and fully exited. This means Chinese capital will control an even larger share of Manus than before Meta's acquisition.
How Manus Works: Technical Architecture Explained
Understanding Manus requires grasping a key conceptual distinction: Manus is not a foundation model company. It doesn't train its own large language models from scratch. Instead, Manus builds sophisticated "agent architecture" on top of existing models from providers like OpenAI, Anthropic, and Chinese AI labs.
The Manus agent system consists of several core components:
1. Cloud Virtual Machine Environment
Manus operates in a sandboxed virtual environment that can execute code, manipulate files, browse websites, and interact with external services. This gives the AI a persistent workspace where it can accumulate information across steps.
2. Task Decomposition Engine
When given a complex goal, Manus breaks it down into executable sub-tasks, creates a visual execution plan, and proceeds step by step. Users can watch the AI's reasoning process and intervene if needed.
3. Multi-Agent Orchestration
Complex tasks are handled by multiple specialized AI agents working together—one might handle web browsing, another data analysis, another document creation. The orchestration layer coordinates their efforts toward the unified goal.
4. Memory and Context Management
Manus maintains context across extended interactions, learning from previous steps and adapting its approach based on intermediate results.
GAIA Benchmark: Beating OpenAI's Research Team
One of Manus's key claims to fame is its performance on the GAIA (General AI Assistants) benchmark, a rigorous test of real-world AI assistant capabilities. GAIA evaluates AI systems on tasks requiring web browsing, file manipulation, multi-step reasoning, and practical problem-solving.
Manus achieved a score of 87.3% on GAIA, outperforming OpenAI's own DeepResearch team, which scored 85.5%. This benchmark victory was part of what attracted Meta's interest and demonstrates that Chinese AI companies can compete at the highest levels of practical AI capability.
The Founders: A New Generation of Chinese Tech Entrepreneurs
Manus's success reflects the rise of a new generation of Chinese tech founders who blend deep technical expertise with sophisticated product thinking.
Xiao Hong (肖弘), the founder and CEO, was born in 1993 and graduated from Huazhong University of Science and Technology. Before Manus, he founded Yiban Assistant (壹伴助手), a B2B SaaS tool for content creators that attracted investment from Tencent. His role focuses on global business development and commercialization strategy.
Ji Yichao (季逸超), the chief scientist, was born in 1992 and leads the core algorithm development that forms Manus's technical moat. His research background in agent architecture and autonomous systems underpins the platform's capabilities.
Zhang Tao (张涛), the product partner, brings experience from ByteDance and Meituan's "Guangnian Zhiwai" AI venture. He oversees both domestic and international product strategy, helping Manus balance its Chinese roots with global appeal.
The HKEX IPO Path: What's Next for Manus
Following the successful buyback, Manus's investors are planning the company's next chapter: a listing on the Hong Kong Stock Exchange (HKEX). The target timeline suggests an IPO filing could come as early as the first half of 2027.
To prepare, Manus is restructuring its corporate architecture. The plan involves creating a new Chinese-registered joint venture to operate domestic business, while maintaining overseas operations through a Cayman Islands holding structure. This dual-track approach satisfies both Chinese regulatory requirements and international investor expectations.
The IPO would come at a potentially fascinating time for AI valuations. Manus's current revenue of $400-500 million annualized, if sustained, could support a significant public market valuation—possibly exceeding the $2 billion Meta paid, given the revenue growth and the company's now-certain independence from American tech giants.
Why This Matters: The Bigger Picture
Manus's story reveals several important dynamics in the global AI race:
1. AI Competition Is Shifting from "Model Wars" to "Agent Wars"
For years, AI competition focused on who could build the most powerful foundation model. Manus represents a shift in focus: from building smarter AI brains to building AI that can actually do things. This productization of AI agents may prove more commercially valuable than raw model improvements.
2. Technology Sovereignty Is Reshaping M&A
The Manus saga demonstrates that AI assets now carry strategic national importance. China blocking Meta's acquisition and Chinese investors quickly organizing a buyback shows that both governments and companies are treating AI leadership as a sovereignty issue. Similar dynamics played out with Nvidia's attempted acquisition of ARM, which was also blocked on national security grounds.
3. Chinese AI Companies Are Learning Productization
Historically, Western observers sometimes criticized Chinese AI companies for focusing too much on implementation rather than fundamental innovation. Manus represents the opposite—a Chinese company succeeding not because of unique model architecture but because of superior product design and user experience. This suggests Chinese AI is evolving beyond "copy and improve" toward genuine innovation.
Challenges Ahead: What Manus Must Overcome
Despite the success, Manus faces real challenges. Technical experts note that current AI agents have reliability issues—multi-step operations can accumulate errors, with some studies showing approximately 12% error rates in intermediate steps. Production-ready reliability for enterprise use cases remains an ongoing engineering challenge.
Additionally, the open-source community is developing competing solutions. Projects like MetaGPT and OpenManus are building similar agent frameworks, potentially eroding Manus's first-mover advantage. The company will need continuous innovation to maintain its technical edge.
Conclusion: The Future of AI Is Agents
Manus represents a pivotal moment in AI history—not because of any single technological breakthrough, but because it demonstrates the commercial viability of AI agents that actually execute tasks rather than just respond to queries. The drama surrounding its acquisition and buyback underscores how strategically important these capabilities have become.
For foreigners trying to understand modern China's tech ambitions, Manus offers a compelling case study. It shows Chinese entrepreneurs who can identify market needs, build sophisticated technical systems, attract global investment, and navigate complex geopolitical currents. Whether Manus ultimately succeeds as a public company or gets disrupted by faster-moving competitors, its story illuminates where AI—and China's role in it—is heading.
The next time someone asks why China matters in AI, you might tell them about Manus: a company that became valuable enough for Meta to pay $2 billion, important enough for Chinese regulators to block a foreign takeover, and promising enough for Chinese investors to buy it back at original prices despite 4-5x revenue growth. In the global AI race, that's the kind of story that speaks for itself.