Meta Cuts Ties With Manus After China Orders Reversal of $2 Billion AI Deal

Meta has started unwinding its $2 billion Manus deal after China’s National Development and Reform Commission ordered the transaction reversed on April 27, 2026. The break from Manus appears to have moved into operations at the start of June, when Manus staff were reportedly blocked from Meta internal systems and Meta employees were told not to use Manus tools for internal projects. By June 11, reports said Meta had completed an operational split from Manus and halted data sharing between the two companies.

That leaves Manus in an awkward public position. The product still presents itself as part of Meta, while the deal behind that message is being pulled apart. Users see the Meta association on the product side, but the company is reportedly being separated from Meta’s systems, internal work, and data access. For an AI agent product that asks people to upload files, connect accounts, create projects, and trust its output, that gap is not a small detail.

The dates matter because the deal did not fall apart in one clean step. Manus went from viral AI agent startup, to Benchmark-backed company, to Singapore-based acquisition target, to Meta-owned product, to Chinese regulatory problem in roughly a year. That sequence makes the acquisition look less like a clean AI win and more like a rushed deal that ran into the exact problems Meta should not have needed to inherit.

How the Meta and Manus Deal Fell Apart

Early 2025

Manus went viral after claiming it had released the world’s first general AI agent, capable of making decisions and executing tasks with less prompting than normal chatbots. The product drew attention because it looked like something beyond a normal chat tool, with the ability to browse, create files, build pages, write code, and work through projects inside a more complete interface.

May 2025

Manus raised $75 million in a round led by Benchmark, helping push the company near the front of the AI agent rush as investors looked for products that could move AI beyond chat and into task execution.

July 2025

Manus shut its China offices, laid off dozens of employees, moved operations to Singapore, and re-incorporated through Butterfly Effect in Singapore. The move made Manus look more global, but it did not stop Chinese regulators from later treating the company as tied to Chinese AI talent, technology, and intellectual property.

December 2025

Meta announced that it would acquire Manus. The deal was reported at roughly $2 billion, with some reports placing the value higher. Meta planned to integrate Manus into products including Meta AI, and Manus began presenting itself publicly as part of Meta while saying the product would continue operating.

January 2026

China began reviewing the sale soon after Meta completed the acquisition. Beijing did not accept the idea that Manus becoming Singapore-based removed the company from Chinese scrutiny.

March 2026

Manus co-founders Xiao Hong and Ji Yichao were summoned to Beijing for talks with regulators and later barred from leaving China. By then, the deal was no longer only a Meta acquisition. It had become a government review involving AI talent, technology, ownership, and control.

April 27, 2026

China’s National Development and Reform Commission ordered the deal unwound. The order prohibited foreign investment in Manus and required the parties to withdraw the acquisition transaction.

May 21, 2026

Manus co-founders Xiao Hong, Ji Yichao, and Zhang Tao were reported to be exploring ways to comply with China’s order, including raising about $1 billion from outside investors to buy the company back. That does not prove Manus is bankrupt or shutting down, but it does show how complicated the reversal became once the company had already been sold.

Start of June 2026

Meta reportedly began the operational break from Manus. Manus staff were blocked from Meta internal systems, and Meta employees were told not to use Manus tools internally.

June 11, 2026

Reports said Meta had completed an operational split from Manus and halted data sharing between the two companies, moving the unwind beyond a regulatory order and into the actual operation of the companies. At the same time, Manus has continued posting small product updates, including a Zoom connector and a Canva connector, releases that look minor next to the scale of losing Meta’s systems, data access, and ownership path. Its blog now reads like a company trying to look normal while the acquisition is being pulled apart, including a deeper collaboration with Similarweb, a public company that remains GAAP unprofitable.

The deal looked confusing from the start because Manus was not a new LLM for Meta to own. Manus is an interface that sits between the user and outside AI systems, including models from companies like Anthropic and OpenAI, then wraps those systems in browser tools, files, code tools, image tools, and task controls. Meta was not buying a new foundation model from Manus. It was buying an interface around other models and tools.

That distinction changes how the acquisition should be read. If Manus had owned a frontier model, unique infrastructure, or a hard technical advantage Meta could not build, the deal would be easier to understand. Instead, Manus was mainly a product layer that organized outside models and tools into a workflow that made AI feel more capable. That can attract users quickly, but it is not the same as owning the model, the compute, or the platform underneath it.

Meta did not need Manus to build that kind of product. Meta already has models, infrastructure, engineers, product teams, browser experience, app distribution, business tools, and enough internal AI work to build a Manus-style agent layer itself. The company has Facebook, Instagram, WhatsApp, Messenger, Meta AI, ads, business messaging, and the engineering base to put agent tools in front of users without buying a product built around other companies’ models.

That makes the deal look more like Meta buying speed during the AI agent rush than buying a core AI asset. Manus had attention because it packaged the right pieces into one product at the right moment. It could browse, write code, create files, generate pages, and make work look like it was moving through a more complete system. That was the part people noticed. The harder question was whether Manus had anything Meta could not create on its own.

The acquisition also gave Manus something it could not give itself. The Meta name made the product look more important, more durable, and more likely to become part of a larger AI platform. A rough agent interface looks different when it is attached to a company with Meta’s scale, engineers, models, and distribution. Without that association, Manus has to stand on the product itself.

That is where the problem becomes harder for Manus. After the acquisition, Manus did not clearly become the stronger product people might have expected under Meta. It did not suddenly become more reliable, more capable, or more useful for real work. From my experience as one of the earliest users, Manus could look impressive while it was working through a task, but it was unusable too often when the result had to be checked, reused, or trusted.

The forced unwind gives Meta a cleaner path than keeping Manus attached. Meta can continue building agent tools inside its own products without carrying a disputed acquisition through Chinese regulatory pressure. It can build the interface, browser layer, file tools, model routing, image tools, workflow system, and business agent features on its own systems. It can keep control of the product layer instead of relying on a company it no longer has a clean path to own.

Manus has the harder road because it has to explain what the company becomes without Meta holding it up. It has to show who controls the product after the unwind, what happens to the team, what happens to user data, how the service will keep operating, and why users should trust Manus as a standalone company after the deal that made it look stronger began falling apart.

The acquisition now reads less like Meta buying a core AI asset and more like Meta buying into the agent rush. Manus arrived early with a polished interface and a strong pitch, but it was not a foundation model, and it was not something Meta could not rebuild. Once China forced the deal apart and Meta began cutting Manus away from its systems, the weakness of the acquisition became easier to see.

Meta is better off without Manus because Meta is not losing a model. It is losing an interface it can rebuild. Manus is losing the company name that made the interface look important, and that is the side of the split with the larger problem.

Manus may continue operating, but the company no longer has the clean story it had when Meta announced the acquisition. The product still exists, but existing is not the same as being a stable platform for serious business work. A company dealing with a forced unwind, stopped data sharing, split operations, possible buyback financing, and an unclear future should not be treated like a settled Meta-backed agent platform.

The deal started as one of the biggest AI agent acquisitions of the boom and is now being pulled apart by regulators. Meta can build the agent layer itself. Manus has to prove that it is more than an interface that arrived during the AI agent rush, got pulled into a massive acquisition, failed to evolve after the deal, and became the center of one of the messiest AI transaction reversals so far.

• Site owners can also find Manus Bot in our bot directory and monitor it through Botcrawl Edge to see when Manus is fetching content from their site and block it when needed. Botcrawl also tracks the broader Meta bot family, so publishers can see when Meta, Manus, or related AI crawlers show up in their traffic.