In March 2024, a general AI agent called Manus quietly launched its beta in China.
No one could have predicted what would happen over the next 18 months: a waitlist of more than 2 million people, annualized revenue reportedly surging to $127M ARR, top-tier global VC Benchmark competing to invest, Meta offering more than $2B to acquire the company, and then the Chinese government halting the deal.
This is not just a financing story. It is one of the most complete and volatile examples so far of a Chinese AI product going global.
What matters here is what Manus got right, and what builders can learn from it.
Strip Away the Hype: What Is Manus?
Most descriptions of Manus use terms such as “general AI agent” or “AI operating system.” Those sound impressive, but they are not specific enough.
In plain language: Manus is an AI that executes tasks for you.
It does not merely give advice and ask you to do the work. You can tell it, “Research prices for 100 pairs of running shoes and organize them into a spreadsheet.” Manus then searches the web, extracts the data, generates a CSV file, and delivers the result.
The technical architecture is substantial. Manus launches a cloud virtual machine for each task, equipped with a browser, code interpreter, office suite, design engine, and other tools. After the user enters a task, the system decomposes the goal, plans the steps, calls tools, executes the work, and delivers a finished output.
The core product insight is this: users are not missing AI advice; they are missing AI execution. Traditional AI assistants follow a “you ask -> it thinks -> you do” pattern. Manus follows “you ask -> it does -> you review.”
That is a fundamental workflow compression.
Three Key Productization Decisions
Manus’s productization path can be unpacked layer by layer.
Layer one: validate demand with a general agent, then expand value with vertical tools.
Manus initially launched a general agent that could understand natural-language task descriptions and execute autonomously. That single broad capability attracted a beta waitlist of more than 2 million people and validated the demand for “AI that does work for me.”
But Manus did not stop there. Later versions introduced AI design, AI slides, image generation, music generation, browser operator features, and other vertical tools. Each tool is a productized packaging of the general agent’s capability in a specific scenario.
The strategy is smart: use “it can do anything” to test market demand, then use “it does this specific job well” to improve completion quality and retention. The general agent is the top of the funnel. Vertical tools become the conversion and retention engine.
Layer two: turn speed into a competitive barrier.
Manus 1.0 had an average task completion time of 15 minutes. For users accustomed to second-level response times, that was slow.
Manus 1.5, released in October 2025, compressed average completion time to under four minutes, while internal task quality improved 15% and user satisfaction improved 6%. The later 1.6 Max version reportedly improved satisfaction by another 19.2%.
In early AI product competition, speed is often underrated as an engineering optimization, but it is actually part of the core product experience. A product that returns an answer in 15 minutes and one that returns it in four minutes feel far more different than the 11-minute gap suggests.
Layer three: use parallel architecture to redefine the perceived ceiling.
The Wide Research feature, released in July 2025, lets users launch up to 100 parallel sub-agents at the same time. That means a user can complete in minutes a research task that might previously have taken a team a week.
Technically, this may be multi-process management plus task orchestration. Product-wise, it is dramatic. It moves the user from “AI helps me do one thing” to “AI does one hundred things for me.”
A Clever Commercialization Design
Manus’s commercialization design is one of the most useful AI product examples from 2024-2025.
Pricing model: credits = flexibility plus ceiling.
Manus uses three pricing layers: Starter at $39/month with 3,900 credits and two concurrent tasks; Pro at $199/month with 19,900 credits and five concurrent tasks; and Team at $39/seat/month, starting at five seats with a pooled 19,500 credits.
The credit model has several advantages:
- For light users: usage is transparent and entry cost is low.
- For heavy users: greater usage naturally pushes upgrades.
- For Manus: credit burn depends on task complexity, so margins can theoretically be managed.
Compared with pricing based on number of conversations or word counts, Manus’s credit model is more flexible. It is closer to AWS-style pricing: users pay for consumed compute resources rather than an abstract number of interactions.
Freemium funnel: 2 million waitlist users -> paid conversion.
During beta, Manus used a free, invite-only model. The effect was twofold:
- It created scarcity. “Not everyone can use it” made people want access more.
- It built a word-of-mouth base. Two million users became natural distribution nodes while they used the product.
When Manus began charging in March 2025, it had already built enough dependency and brand awareness. First make users unable to live without you, then discuss price.
Hidden globalization revenue: enterprise partnerships.
In November 2025, Manus became a launch partner for Microsoft Agent 365. That means Manus workflows can run natively inside the Microsoft 365 ecosystem.
For Manus, this turns Microsoft’s enterprise sales motion into a distribution channel. Turning a potential competitor’s ecosystem into your own distribution network is a classic coopetition move.
The Growth Code
Several data points stand out in Manus’s growth curve:
- Beta period, March 2024 to March 2025: 2 million waitlist users, zero revenue
- Eight months after paid launch: $100M ARR
- End of 2025: $127M ARR, with a reported run rate of $125M
- Monthly growth rate: more than 20%
That growth is top-tier among AI products. What explains it?
First, product output is inherently shareable.
When Manus completes a complex task, such as researching 100 running shoes and building an analysis table, the user does not receive only a chat response. They receive a finished artifact. That artifact itself is content. It can be screenshotted, shared, and reposted.
Compared with ChatGPT conversation screenshots, the impact of “AI completed a whole piece of work for me” is stronger. Great AI products do not only make users think, “AI is smart.” They make users feel, “I am incredibly productive.”
Second, the waitlist created FOMO.
Two million people waiting for an invitation is itself a market signal. When prospective users see that many people waiting, their first instinct is often not skepticism; it is “I want to try it too.”
Third, globalization opened the ceiling.
In 2025, Manus moved its headquarters from China to Singapore and opened offices in Tokyo and San Francisco. At the time, this may have looked aggressive. In hindsight, given the later Meta acquisition and regulatory halt, it looks farsighted.
Meta’s $2B Acquisition Attempt and China’s Halt
In December 2025, Meta announced a deal to acquire Manus for more than $2B, making it one of Meta’s largest AI acquisitions.
In May 2026, however, Chinese authorities halted the transaction on national security grounds. Reports suggested that regulators viewed Manus’s technology as involving sensitive “AI core technology export.”
The story then became even more dramatic. Manus’s founding team reportedly initiated a $1B buyback offer to regain control from Meta. As of the end of May 2026, the final outcome remained unresolved.
This creates major uncertainty. But even setting politics aside, the product value of Manus had already been validated. Meta would not have offered $2B, Benchmark would not have led a $75M round, and 2 million users would not have waited in line if there were no real product pull.
What Can Be Copied and What Cannot
Moves builders can copy:
- General -> vertical product matrix. Use a general agent to quickly validate demand for “AI that does work,” then develop vertical tools for high-frequency tasks such as design, slides, images, and music.
- Credit-based pricing. It is more flexible than fixed interaction limits and more reasonable than word-based pricing. Light users can start cheaply, while heavy users have a clear upgrade path.
- Waitlist plus free strategy. In early stages, do not rush monetization. Scarcity and word of mouth can build a user base before monetization begins.
- Global compliance early. Moving headquarters to Singapore and distributing offices reduced risk before geopolitics became an immediate issue.
- Borrow ecosystem power. A Microsoft Agent 365 partnership turns a potential competitor’s channel into a distribution partner.
First-mover advantages and luck that are hard to copy:
- The general-agent time window. In early 2024, very few general agent products existed. Manus caught that open window.
- Top VC willingness. Benchmark investing in a Chinese AI company in 2025 may have been a window that is harder to reopen under geopolitical pressure.
- Global exposure from Meta’s offer. Even though the deal did not complete, Meta’s reported $2B offer gave Manus a level of brand validation that money cannot easily buy.
- Technical depth. Optimizing Claude models, building proprietary context engineering, and managing large-scale parallel agent architecture are not capabilities that can be copied quickly with capital alone.
What to Watch
The outcome of the Meta-Manus transaction will deeply influence Manus’s future. But leaving that uncertainty aside, Manus has already shown a clear direction for AI productization:
In the second half of AI product competition, the winner may not be the company with the strongest model. It may be the company that packages AI execution into the easiest product.
Manus proved in two years that users will pay when AI moves from “advisor” to “executor.”
If you are building an AI product, ask yourself: is my product giving users advice, or doing the work for them?
The answer may decide how far the product can go.
Data note: Revenue and financing figures in this article are based on Sacra’s public company profile for Manus. Transaction information is based on Reuters, TechCrunch, CNBC, Fortune, and other mainstream media reports. Product pricing and feature information comes from Manus’s official website, manus.im. Revenue figures are Sacra estimates based on public information and do not constitute financial advice.