The archive · AI & Models · Strategic decision · 2022–2026
Manus's general-agent bet: viral launch, $2B+ Meta buyout — then regulators unwound it
A general agent that does tasks, not a chatbot that talks: Manus went viral, hit ~$500M, sold to Meta for $2B+, then regulators unwound it.
Manus · Butterfly Effect
What the business is
Manus is a general AI agent service that carries out multi-step tasks — screening résumés, planning trips, building websites, analysing stocks — by driving its own virtual computer in the cloud, sold as a subscription.
Starting capital：>$10M from Tencent, ZhenFund and Sequoia China, then a $75M Benchmark-led round in April 2025 at roughly a $500M valuation
How it started
Xiao Hong, a serial founder who had built WeChat-ecosystem tools, founded Butterfly Effect in Beijing in 2022, first shipping the Monica browser assistant. In March 2025 it launched Manus with a demo of a 'general AI agent' that screens résumés, plans trips and analyses stocks; the demo went viral, more than 3.5 million people applied for invitation codes, and the startup was compared to DeepSeek as a Chinese AI surprise.
What happened
In April 2025 Benchmark led a $75M round at roughly a $500M valuation, and Manus began charging $39/month with a $199 tier. From mid-2025 the company cut staff in Beijing and Wuhan, moved its core team to Singapore, stopped serving Chinese users and withdrew from Chinese social media — a relocation reportedly spurred by a US outbound-investment probe into Benchmark. In December 2025 Meta agreed to buy the company for a reported $2B+, Meta's third-largest deal ever, with Manus to keep operating independently and Xiao to join as a Meta vice president; the service had passed $100M annual recurring revenue after processing 147 trillion tokens.
How it ended up
In April 2026 China's top economic-planning body prohibited the acquisition and ordered the parties to withdraw; in June reports said Meta had cut Manus off its internal systems; on August 11, 2026 Manus confirmed it would 'soon return to operating as an independent company.'
Background
Butterfly Effect was founded in Beijing in 2022 by serial entrepreneur Xiao Hong, who had previously built WeChat-ecosystem tools. Its first product was Monica, a browser assistant; in March 2025 it launched Manus, a 'general AI agent' that completes multi-step tasks — screening résumés, planning trips, building websites, analysing stocks — by driving its own virtual computer in the cloud rather than just chatting. A slick demo went viral, more than 3.5 million people applied for invitation codes, and codes resold for thousands of dollars.
Benchmark led a $75M round in April 2025 at roughly a $500M valuation, on top of earlier backing from Tencent, ZhenFund and Sequoia China; Manus started charging $39/month. From mid-2025 the company laid off staff in Beijing and Wuhan, moved its core team to Singapore, stopped serving Chinese users and left Chinese social media — a relocation reportedly triggered by a US outbound-investment probe into Benchmark, and read as an attempt to distance itself from Beijing.
In December 2025 Meta agreed to buy Manus for a reported $2B+, its third-largest acquisition ever, with Manus to run independently and Xiao to become a Meta vice president; at the time the service had processed 147 trillion tokens and passed $100M in annual recurring revenue. The deal fell apart: in April 2026 China's top economic-planning body prohibited it and ordered withdrawal, and on August 11, 2026 Manus confirmed it would return to operating independently.
As of September 2026 Manus is back as an independent company, its $2B+ exit unwound. The episode turned a product win — one of the fastest-scaling AI agent launches of 2025 — into a cautionary tale about what a startup's ownership geography means when home regulators and foreign acquirers collide.
What has to be true
- The bet was on a product category, not a moat: Manus wagered that autonomy — an agent doing the work in a cloud computer — would beat chat, and its demo proved the category in days.
- Scarcity was the wedge: invitation-only access made demand visible (3.5M+ applications), turning virality into pricing power and a fast subscription business.
- The exit bet ignored a second stakeholder: by moving to Singapore and selling to Meta, Manus treated global users and US capital as the whole boardroom — its home regulator held the deciding vote.
- The collapse wasn't operational: the product kept working and ARR passed $100M; the deal died on regulatory review, which is why the lesson is about geography and ownership, not product-market fit.
What can be applied
Product virality can win users and an exit, but when home regulators and foreign acquirers disagree, geography and ownership are part of the business plan — not afterthoughts.
Aftermath
As of September 2026, Manus is returning to independent operation after China's top economic-planning body prohibited Meta's reported $2B+ acquisition in April 2026 and ordered withdrawal; Beijing had earlier restricted two Singapore-based co-founders from leaving China, and Meta reportedly cut Manus off its internal systems in June. The product never shut down — subscriptions continued. Founded in Beijing in 2022 and moved to Singapore in mid-2025, the company now faces the harder version of its bet: staying global while its home regulator has shown it can veto the ownership structure.
Sources
- Chinese AI startup Manus scores funding at $500 million value
- Meta to Buy AI Startup Manus in a Deal Reported to Be Worth Over USD2 Billion
- Caught between great powers: the cautionary tale of Manus
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card