Manus Closes Record $500M Round at $4 Billion After Meta Deal Collapsed

Butterfly Effect, the Singapore-based parent of AI agent platform Manus, has confirmed it raised more than $500 million in a new financing round led by Boyu Capital and IDG Capital, with existing investors Tencent, Sequoia China (HSG), and ZhenFund continuing as backers. The round is the largest single raise in China's AI application sector on record, and it arrives less than two months after Butterfly Effect formally resumed independent operations following the collapse of Meta's $2 billion acquisition bid.

The company did not disclose a final valuation. Bloomberg reported in mid-September that the round was approaching completion at a target of approximately $4 billion — roughly double the implied price when Manus and its founders bought their equity back from Meta earlier this year.

A Startup That Survived the Acquisition That Was Supposed to End Independent AI

Manus launched in March 2025 as an invitation-only autonomous agent, capable of handling tasks — resume screening, stock research, travel planning — with minimal human direction. It drew over a million views on its demo within 20 hours of launch, and invitation codes, scarce by design, sold on secondary markets for as much as $13,800, according to reporting by China Daily. By the end of 2025, the company reported a revenue run rate around $125 million annually.

What followed the initial success was one of the more turbulent corporate trajectories in recent AI history. Meta acquired Butterfly Effect in December 2025 for a reported $2–3 billion. China's Ministry of Commerce launched an evaluative investigation in January 2026, citing export-control and technology-transfer concerns. On April 27, the National Development and Reform Commission blocked the deal outright and ordered both parties to withdraw. Meta announced it was severing ties in June. By August, Manus was formally transitioning back to independent operations, completing the process later that month.

The $500 million close, announced on October 8, is the first external financing Butterfly Effect has disclosed since regaining autonomy.

What the Money Is Backing: Cascade, Cue, and a Domestic China Product

Ten days before the funding announcement, Manus released its 2.0 version and a new standalone product called Cue. The product update helps explain why investors chose this moment to commit capital at scale.

The centerpiece of Manus 2.0 is Cascade, the company's in-house agent orchestration framework. Rather than loading every available tool into an agent's working context for every task — a design that inflates token consumption and slows execution — Cascade activates only the capabilities a specific task requires, loading them dynamically as the job progresses. The result, according to Manus's own benchmarks from a single tested configuration, is about 23% lower token usage, tasks completing 28% faster, and approximately 32% lower per-run operating cost compared with its prior system. Those figures are company-reported and have not been independently reproduced, and Manus has not published the task mix, model configuration, or failure-rate data behind them.

Cascade's design resembles a pattern better known at the model layer — mixture-of-experts architectures selectively activate subsets of parameters per token rather than running the whole network every time — applied here at the orchestration layer. The analogy is imperfect, but the underlying logic is the same: idle capacity costs nothing if it never runs.

Cue, the companion product built on the same infrastructure, takes a different architectural bet. Rather than treating an AI agent as a cursor moving inside the user's own accounts, Cue gives each agent its own durable identity: a dedicated email address, a phone number, a budget-constrained spending wallet, and a persistent cloud-hosted computer with a public IP address. Agents can accept inbound phone calls on behalf of the user, send messages, make purchases within user-defined spending limits, and hand work off to other agents in shared group contexts.

The Cloud Computer underlying Cue's persistent environment runs Ubuntu Server 24.04 LTS. Current pricing starts at $30 per month for the Standard plan (2 vCPUs, 4 GB of RAM, 70 GB of storage), scaling to $50 per month for the Advanced plan (2 vCPUs, 8 GB of RAM, 120 GB of storage). Event-driven automations — triggering on new emails, Slack messages, calendar entries, or Notion updates — replaced the prior schedule-only approach.

According to the Silicon Stars report of the funding announcement, Manus is currently in active development of a domestic China product, with official launch described as "not far away." The company is recruiting across 17 roles, including agent product managers and Harness engineers — a signal that Cascade's orchestration layer is being treated as a platform to build on, not a one-time product feature.

Independent in a Market Where Big Tech Controls the On-Ramps

The practical obstacle facing Manus in its home market is distribution. Since early 2026, Tencent, Alibaba, and ByteDance have each consolidated their internal AI agent projects under a single branded umbrella. WorkBuddy (Tencent) reported more than 20 million monthly PC-side visits in June 2026, ranking first among domestic AI office agents according to data from Analysys, up from about 8.85 million at its launch. Qianwen Office (Alibaba) integrates bidirectionally with DingTalk, covering messages, documents, schedules, and approvals. Doubao Work (ByteDance) draws context from Feishu, offers a Windows virtual desktop for isolated long-running tasks, and provides more than 100 specialized "work partners" for multi-agent collaboration.

Each of these platforms gains most of its distribution through messaging and productivity ecosystems the respective company already controls. WorkBuddy's core logic originated in Tencent's CodeBuddy programming tool; the others piggyback on office suites with hundreds of millions of daily active users. For an agent that relies on a company's collaboration platform to reach users, building a technically better product may matter less than which office suite a company has already standardized on.

Manus and Cue are designed around a different premise: platform independence. The products are browser-first and do not require users to already live inside a particular app ecosystem. That independence is also the reason Tencent — which leads the competition domestically — is simultaneously Manus's largest individual external shareholder. Tencent's position in the cap table predates the current competitive configuration, but it creates a structural ambiguity the funding round does not resolve.

On the global stage, Meta's own personal agent, Muse, launched September 8 — three weeks before Manus 2.0 — distributing through WhatsApp and Meta's app using the company's proprietary Muse Spark model. Muse can navigate websites, fill forms, and make Stripe-backed payments, and Meta has acknowledged that prompt injection — an attack where malicious content in a page hijacks an agent's next action — remains an unsolved problem in the field. That acknowledgment applies equally to Manus: the attack surface Cue opens by giving agents real phone numbers and payment capability is not a hypothetical, and no public technical documentation describes how spending limits are enforced at the protocol level.

Read more: Instinct raises $1B at $10B valuation as personal AI agent competition intensifies

Why $500 Million for a Platform That Doesn't Train Its Own Models

The Manus funding round sits within a broader argument about where value will accumulate as AI matures. Manus does not train its own foundation models — a fact that emerged publicly when the U.S. Treasury reviewed Benchmark Capital's Series B investment in early 2025 and concluded the transaction fell outside Outbound Investment Security Program requirements, in part because Butterfly Effect was a model orchestrator rather than a foundation model developer, and because it was incorporated in the Cayman Islands.

That determination, intended as a regulatory line, doubles as an investment thesis. If foundation models continue commoditizing — with capable open-weight models reducing the barrier to capability — then the differentiation moves upward: to the orchestration layer that determines which tools get loaded, in what sequence, at what cost; to the product layer that determines what users will pay for and trust with their email, their phone, and their money; and to the data and feedback layer that improves the system as agents complete more tasks.

The $500 million round, led by Boyu Capital and IDG Capital alongside Manus's original backers, is a bet that Butterfly Effect occupies a durable position in that stack — and that surviving the Meta acquisition attempt, the NDRC block, and the forced restructuring leaves it positioned to be the best-funded independent challenger in a market that big tech would otherwise consolidate entirely.

Read more: Manus 2.0 launches Cascade framework and Cue personal agent