Why Tencent Prefers Acquiring Meta's Manus Over Incubating Its Own

I noticed an interesting detail: After the news broke, Tencent's HK stocks dropped by 2.21%. The market voted with its feet, indicating that most people think this deal by Tencent isn't worth it.

Photo by Klub Boks / Pexels

But thinking carefully, from an engineering perspective, Tencent's move is actually quite pragmatic.

First, let's look at what Manus is. As a star project in the AI Agent track, Manus is essentially an "AI workflow orchestration platform." It integrates large models, toolchains, and API interfaces, allowing users to build complex automated tasks using natural language. The tech stack isn't particularly deep, but the engineering implementation is solid, especially in task decomposition and tool calling, which is more mature than many contemporaneous Agent frameworks.

Why is Meta selling it? From Meta's public information, there is significant internal disagreement regarding the Agent direction. Zuckerberg is betting on the metaverse and open-source large models; Agents, which lean towards B-end and toolchains, don't fit well with Meta's C-end social DNA. Moreover, Manus operates independently in Singapore, resulting in high management costs for Meta. It's better to cash out and use the money to invest in Llama 4 or VR hardware.

Now back to Tencent. Tencent has Cloud, WeChat Mini Program ecosystem, and WeCom, but it has always lacked an "automation brain" that connects these products. Honestly, Tencent's own Hunyuan large model is okay for C-end dialogue scenarios, but in terms of Agent orchestration, it lags significantly behind Manus.

[!note] There is a consensus in the tech circle: Large models are just engines; Agent frameworks are the transmissions. What Tencent lacks isn't the engine, but the transmission.

If Tencent were to build an Agent platform from scratch, it would take at least 1.5 to 2 years. Along the way, they would step into countless pitfalls: logical robustness of task decomposition, error handling for tool calls, context management for multi-turn conversations... Manus has already navigated these engineering details. Buying a mature product line for $2 billion actually saves Tencent time costs.

Looking at valuation again. $2 billion, calculated based on Manus's revenue scale last year, puts the Price-to-Sales (PS) ratio around 50x. For a SaaS product, this is indeed expensive, but considering this is buying out a mature Agent platform and Tencent doesn't need to rebuild the team, the price is actually reasonable.

Sequoia and ZhenFund's willingness to co-invest indicates they also see long-term value in the Agent track in the Chinese market. Three institutions taking over together shares the risk, so the burden on each isn't too heavy.

For the Manus team, independent operation in Singapore is a good arrangement. Tencent not holding a controlling stake means the team can maintain its engineer culture and won't become bloated due to big-company management processes. This is a typical "risk avoidance" strategy in AI startups—many big companies acquire AI firms, only to drag them to death with internal bureaucracy.

However, there are three key issues worth watching:

First, Manus's current main customers are concentrated overseas. Can the Singapore team's compliance capabilities and data processing capabilities adapt to domestic enterprise service scenarios? Tencent's WeChat ecosystem has a massive number of SMEs and developers whose demand for Agents is fragmented. Whether Manus's existing product can quickly adapt is a test.

Second, how deep is the integration between Tencent's Cloud services/WeCom and Manus? If it's just simple API docking, then spending $2 billion seems wasteful. The real value lies in embedding Manus's Agent framework into Tencent's entire product matrix, allowing users to invoke various Tencent Cloud services using natural language within WeCom. That would be a killer application.

Third, after Meta exits, will Tencent influence Manus team's technical roadmap? Tencent's style in AI is conservative, while Manus's core competitiveness is rapid iteration and engineering innovation. If Tencent sends a lot of product managers in to "integrate," they might turn the product into a chimera.

From a purely technical perspective, the core logic of this deal is: Tencent bought an entrance to an "AI middleware" for $2 billion. In the coming years, the core battlefield for enterprise-level AI applications will shift from the large models themselves to Agent orchestration platforms. Whoever controls this middleware will hold the initiative in the B-end market.

Meta's exit indirectly confirms: General-purpose large models and Agent platforms are two different tracks. There are no absolute technical barriers, only engineering implementation capabilities. Tencent's choice is very pragmatic—rather than reinventing the wheel, buy one that's already rolling.

[!tip] Developers should also pay attention to this news: After Tencent takes over Manus, it may open up more API interfaces and plugin ecosystems. If Tencent really wants to scale this platform, it will definitely lower the barrier for developers, allowing more people to build their own Agent applications based on Manus. This is good news for the domestic AI developer ecosystem.

Finally, a trend prediction: In the next six months, Tencent will likely integrate Manus into WeCom and Tencent Cloud, launching a set of "Enterprise-Level Agent Solutions." If this product works, the landscape of the domestic Agent track will change completely.

Original link: https://www.tmtpost.com/8062355.html