On August 11, 2026, AI agent brand Manus released an open letter to its users, formally announcing its resumption of operations as an independent company. This comes seven and a half months after Meta's high-profile announcement of its acquisition of Manus' parent company, Butterfly Effect, for more than $2 billion.
From its explosive debut in March 2025 with invitation codes that were nearly impossible to obtain, to relocating its headquarters to Singapore and severing ties with the Chinese market in June, to the lightning-fast acquisition by Meta in December, followed by regulatory intervention from Chinese authorities, and now a return to independence—Manus has completed a full dramatic cycle of product explosion, capital frenzy, regulatory reversal, and a return to square one in just over a year. This butterfly that once stirred a massive storm in the AI industry, after struggling to flap its wings for seven months amid the turbulence, has finally reached its moment of independence.
From a $2 Billion Acquisition to Regulatory Halt: A Compliance Masterclass in Cross-Border AI M&A
To understand the significance of Manus' resumption of independent operations, one must first trace the full arc of what has been called the most dramatic acquisition in AI history. In March 2025, startup Butterfly Effect released an early preview of its AI assistant Manus, calling it the world's first general-purpose agent product. Users only needed to give Manus a simple instruction, and it would automatically handle complex tasks such as data collection, data analysis, coding, and presentation creation in the cloud. A viral demonstration video ignited the tech world within 48 hours, with invitation codes being resold on secondary platforms for tens of thousands of yuan, and the waitlist surpassing 2.6 million people by the end of March.
The pace of commercialization was equally astonishing. Just eight months after launch, Manus' Annual Recurring Revenue (ARR) broke through the $100 million mark, making it the fastest startup in global history to go from zero to $100 million in ARR. This growth trajectory directly captured the attention of Meta founder Mark Zuckerberg. In December 2025, Meta completed the $2 billion acquisition of Manus in just 10 days, marking Meta's third-largest acquisition ever. However, this seemingly perfect entrepreneurial fairy tale soon hit headwinds.
Manus' origins dictated that this acquisition was far from an ordinary business transaction—its early research and development was primarily conducted in China, and its technical team consisted of Chinese engineers. These key characteristics meant that the flow of its personnel, technology, and data was inevitably tied to Chinese interests. Industry lawyers pointed out that the Manus acquisition involved transferring AI business assets from within China to overseas entities and ultimately selling them to a foreign company, Meta, which triggered compliance risks under cross-border investment regulations. On January 8, 2026, the Ministry of Commerce responded at a regular press conference that it would, together with relevant departments, conduct an assessment and investigation into the consistency of this acquisition with laws and regulations related to export controls, technology import and export, and outbound investment.
On April 27, the Office of the Security Review Mechanism for Foreign Investment (National Development and Reform Commission) lawfully issued a prohibition decision on the foreign acquisition of the Manus project, requiring the parties to rescind the acquisition transaction. This made Manus the first foreign investment acquisition project in the AI industry to be publicly halted since the implementation of the Measures for the Security Review of Foreign Investment in 2021. Following the ban, Manus and Meta began the process of operational separation. According to earlier reports, by early June, Meta had completed the isolation of data, business operations, and underlying algorithms from Manus, thoroughly cutting off all data exchange channels between the two parties.
Meanwhile, Manus' founding team—Xiao Hong, Ji Yichao, and Zhang Tao—began preparing a new funding round of approximately $1 billion to buy back the company from Meta, at a valuation no lower than the $2 billion Meta had paid. According to multiple media reports, in July 2026, Tencent was leading the formation of a Chinese capital consortium, negotiating to repurchase all of Manus' equity from Meta at a valuation of approximately $2 billion. After the buyback, Tencent would become Manus' largest single shareholder while maintaining a minority equity position, with Manus continuing to operate independently in Singapore. The August 11 open letter marks the final execution phase of this complex series of capital maneuvers.
As part of resuming independent operations, and to comply with regulatory requirements in certain jurisdictions, some user data generated on or after December 29, 2025 (the date of the Meta acquisition) will be deleted between August 23 and 24, 2026. Affected users can back up their data through official tools before August 23, with data restoration resuming on August 25. Manus specifically emphasized that this adjustment is unrelated to any data breach or security incident.
The reversal of this acquisition reveals a profound reality: for technology companies with global ambitions, technological innovation determines how fast you can run, commercial capability determines how firmly you can stand, and compliance capability, to a large extent, determines how far you can go. Manus' experience serves as a vivid footnote to this assertion.
After Returning to Independence: The General Agent Arena Is No Longer That Blank Slate
The news of Manus resuming independent operations is certainly noteworthy, but an unavoidable question looms: after leaving the Meta ecosystem and enduring nearly eight months of capital turbulence, can Manus still replicate the explosive momentum of March 2025? The answer is far from certain.
The most fundamental change is that the competitive landscape for general-purpose agents has been completely transformed. When Manus debuted in March 2025, the market was indeed in a blank period for general agents. At that time, the industry's primary concern was whether an agent could actually complete a complex task autonomously from start to finish. Manus seized first-mover advantage with a complete general agent product. But by August 2026, OpenAI, Anthropic, and Google are integrating agents into their core products, while startups like Genspark and OpenCode are also pushing forward from different directions. Chinese tech giants like ByteDance and Baidu are accelerating their overseas product deployments as well. The primary entry point for general large model applications is shifting from chatbots to agents. After regaining independence, Manus faces a far more difficult race than before.
Even more challenging is the impact of open source. Open-source projects represented by OpenClaw are rapidly replacing closed-source paid agent products, and the AI agent sector is shifting from a closed-source paid model toward an open-source universal access direction. Manus' business model—a paid general agent built on closed-source large model orchestration—is facing direct competition from free open-source alternatives.
However, Manus is not without cards to play. Over the past seven months, despite the external turmoil, Manus' product iteration has never stopped. The official blog has published more than 30 product updates, from My Computer and Cloud Computer to Projects, Skills, Branch, and Plan Mode, maintaining a remarkably high update frequency. Since July, Manus has released four additional features: conversation Branch, Auto-Publish for websites, intelligent PPT generation, and Plan Mode. These features address the problems that arise when agents are used over extended periods—whether materials from months-long projects remain usable, and whether the workflows a team has established need to be re-explained each time.
From a commercial data perspective, Manus' revenue actually increased rather than declined during its six months of integration with Meta. Based on mobile platform revenue calculations, Manus' revenue in May 2026 was three times its level at the end of 2025, reaching $3.6 million. In June, The Information reported that its annualized revenue run rate had reached $400 million to $500 million. This indicates that market demand for the product remains robust. However, without Meta's global traffic and ecosystem support, whether Manus can maintain this growth trajectory remains uncertain.
Furthermore, Manus still cannot provide services in mainland China, meaning it has lost access to the world's largest single market. There were earlier reports that the team was collaborating with Alibaba on a mainland version, but with the headquarters relocation to Singapore and the disbanding of the domestic team, that collaboration appears to have quietly fizzled out.
From a capital perspective, Tencent's entry provides important strategic backing for Manus. As an early shareholder, Tencent had already invested in Manus' Series A round at a $85 million valuation, and had realized a return of more than 23 times at the time of the Meta acquisition. In leading this buyback, Tencent values Manus' scarcity as the world's first truly operational general-purpose agent product. However, Tencent's position is that of the largest single shareholder rather than a controlling entity; Manus will continue to operate independently in Singapore, with the potential for an independent public listing in the future. This model of capital support without operational interference provides Manus with financial security while preserving its entrepreneurial flexibility.
For Manus, the strategic choices after returning to independence are crucial. Should it continue to compete through product insights, or invest in more core technologies such as model training to face the most brutal challenges of large model applications? Given Manus' established technical approach—hybrid orchestration of various mainstream large models rather than proprietary development—the former may be the more realistic choice. But whether product-level innovation alone can sustain a differentiated advantage amid the dual pressures of tech giants and open source remains a significant question mark.
Manus' story encapsulates the speed and complexity of this wave of AI entrepreneurship. It completed the full trajectory from product explosion to global expansion to acquisition by a tech giant in an extremely short period, only to be forced back to square one by regulatory review. Now independent once again, it faces a general-purpose agent market with more players, fiercer competition, and clearer rules. Whether this butterfly that once stirred a storm can flap its wings again after the winds have settled—the answer may lie in the series of new features it has promised to unveil next.