For most of 2026, a single Chinese-built software agent seemed to be slipping out of reach of its own founders. A blockbuster sale to an American technology giant had been announced, then abruptly unwound by regulators on both sides of a widening divide.
Now the same product is back in Chinese hands — and investors are writing one of the largest checks ever handed to an artificial-intelligence application maker. The vote of confidence arrives barely a month after the company said it had resumed independent operations.
Manus is not a model lab. It is an agent (智能体): software that opens a computer, clicks through tasks, and delivers a finished file rather than a chat reply. That distinction sits at the heart of why so much money is now chasing it.
From a blocked sale to a record round
The backstory reads like a geopolitical thriller. In December 2025, Meta agreed to acquire Manus for about US$2.5 billion (≈ RMB 17.75bn / HK$19.5bn), after the startup reported annualized revenue above US$100 million and had moved much of its staff to Singapore. Then in April 2026 China's National Development and Reform Commission prohibited foreign investment in the project on national-security grounds, citing the flow of artificial-intelligence technology, data and talent to a geopolitical rival. The deal was unwound; early backers bought the shares back at roughly a US$2 billion (≈ RMB 14.2bn / HK$15.6bn) valuation; and by May the two sides had halted data sharing.
On 1 September 2026, Manus announced it had resumed independent operations, with founder Xiao Hong (肖弘) and the original team back in control. Just over a month later, on 8 October, the parent company Butterfly Effect said it had closed a funding round of more than US$500 million (≈ RMB 3.55bn / HK$3.9bn).
The money and the names
The round was led by Boyu Capital and IDG Capital, with earlier shareholders Tencent, HongShan (formerly Sequoia China) and ZhenFund following on. Bloomberg and the Wall Street Journal, reported by Yahoo Finance and Axios, pegged the post-money valuation at about US$4 billion (≈ RMB 28.4bn / HK$31.2bn) — roughly double the price at which insiders bought the company back from Meta.
That makes Manus one of the most richly valued artificial-intelligence agent makers anywhere by private-market standards. The speed is the headline: a company written off by some as a casualty of the Meta saga instead turned the forced breakup into a fundraising advantage.
Why an application layer still commands billions
Skeptics have long argued that improving foundation models will simply absorb agent startups — why pay for a wrapper when the underlying model already performs the task? The Manus round is evidence that serious investors disagree, at least for teams with brand, traction and a regional moat.
Manus positions itself as a "general agent," analogous to selling a computer rather than a single app. On 28 September it launched Manus 2.0, built on a new in-house execution system, and Cue, a standalone personal assistant that gives each agent its own email address, phone number and mobile wallet. A domestic Chinese version is in development with local model partners. The narrative for backers is distribution: in the world's largest mobile-internet market, owning the user relationship at the agent layer can outlast any single model's pricing war.
The geopolitical pivot
Manus also illustrates a structural shift for China-linked artificial intelligence. With Western acquisition routes now carrying real regulatory risk, domestic and regional capital has become the realistic path to liquidity. Bloomberg reported the company is considering a restructuring to support a Hong Kong listing, which would keep its capital-markets activity in a Beijing-friendly jurisdiction. Tencent's continued participation signals strategic alignment, not merely financial return.
The implication for global operators is direct. China's most visible agent startup is no longer a candidate for a Silicon Valley exit; it is a bet on Beijing-and-region capital, an independent product roadmap, and a market that rewards application-layer reach over raw model scale.
Honest limitations
The round size and valuation are company-reported and corroborated by Bloomberg and the Wall Street Journal, but Butterfly Effect did not disclose detailed revenue or profit figures, so the "most valuable agent maker" label rests on private metrics outsiders cannot check. A core risk remains that broader foundation models — from OpenAI, Anthropic and domestic labs such as Moonshot's Kimi — keep adding agentic features that erode a standalone product's edge. The national-security block on the Meta deal shows the regulatory overhang is not theoretical, and any Hong Kong listing remains a plan rather than a filed certainty. Finally, large funding rounds price future promise, not proven, durable profitability.
What readers can do now
- Watch for an eventual Hong Kong filing if the reported restructuring proceeds, since that would be the first public validation of agent-economics rather than model-economics.
- Track whether application-layer agents retain pricing power as foundation models ship their own computer-use and task-execution features.
- For anyone assessing China-linked artificial intelligence, treat Manus as the clearest signal that domestic and regional capital — not Western acquirers — is now the default liquidity path.