China Telecom leads Maniformer, naming the carrier stack as part of the embodied-AI data product

Six months after founding, Maniformer closed a round in the hundreds of millions of yuan (roughly $30M to $70M at recent exchange rates) led by China Telecom, with Zhangjiang Group, Sequoia China and Yuanqi Innovation participating. The investor list reads like a routine AI funding announcement. The mechanism is different. State-strategic capital is reclassifying physical-AI training data as carrier-grade infrastructure, with the telecom stack named as part of the data product rather than a separate investor relationship.

Maniformer, founded in early 2026, builds the data layer for embodied AI: the training material that lets robots see, move and interact in the real world. The round, first reported by Lei Feng Net on August 17, is one of the fastest ascents by a Chinese embodied-AI data company and the first time a state-owned telecom carrier has led a round in the category.

The structural signal sits in the wording. China Telecom is not described as a financial investor writing a check. The press language, repeated across Lei Feng Net, NetEase aggregations and Pedaily, names cloud-network convergence, compute scheduling, scenario access, and security compliance as part of a "cloud-network + embodied data" co-built product. Zhangjiang Group, a Shanghai state-owned platform, contributes industrial parks and commercial scenarios, positioning the deal as a data-element utility that other embodied-AI companies can plug into. Maniformer's chairman and CEO Yao Maoqing described the round, in a Lei Feng Net exclusive, as "not just capital but cloud, scenario, and ecosystem synergy."

That language, if reflected in the deal structure rather than the press release, recasts the round. Physical-AI training data, the layer most embodied-AI teams describe as their binding constraint, is being treated as carrier-grade infrastructure on the same shelf as cloud, network and compute. The lead investor is positioning to control the layer that decides who gets data and on what terms.

The investor composition supports that thesis. China Telecom brings state-strategic cloud and connectivity. Zhangjiang brings a Shanghai SOE platform with scenario access. Sequoia China, one of the country's two or three top-tier venture firms, over-subscribed as an existing shareholder. Each leg covers a different risk: government alignment, scenario supply, and competitive market validation. For a six-month-old company, that combination is the story. The check size is the second-order detail.

The risks for that thesis are spelled out below. Maniformer's headline metrics are all self-reported. The company says its MEgo hardware, a body-free data-collection rig, has been deployed in factories, logistics, retail, hospitality, security and eldercare scenarios with millimeter-level trajectory precision. The company says its MEgo Engine data-governance platform runs a 10x efficiency lift over traditional annotation and lifts task success 5-10 percent per closed loop. The company says it has built innovation centers in 20-plus Chinese cities and overseas nodes in 5-plus countries. None of these numbers have been independently verified, and the press language, "deployed" rather than "contracted" and "tens of millions of hours of capacity" rather than delivered hours, leaves room between marketing and revenue. "Deployed" can mean a pilot. "Tens of millions of hours" can mean installed capacity, not produced output.

A second risk is single-customer exposure. Maniformer says it is supplying training data to Ant Lingbo's LingBot-VLA 2.0 foundation model. The relationship is a single anchor in a single client. If Ant Lingbo scales, builds its own data stack, or pulls back, Maniformer loses the proof point. The press language does not address how the relationship is contracted or whether it is exclusive.

A third risk is deal structure. "Lead investor" in a state-strategic round can mean a direct equity check from a corporate venture arm, a co-investment with a state fund, or a check-plus-purchase-order bundle that mixes equity, cloud commitments, and scenario access. Lei Feng Net does not name the China Telecom vehicle or the equity-versus-order split. Until that is on the record, the carrier-grade infrastructure interpretation is a working hypothesis, not a closed case.

The falsifier is also clear. If the next round reverts to standard venture timing, twelve to eighteen months out with no SOE lead and no carrier-stack language, the carrier-grade infrastructure thesis collapses and the round looks like a hot deal that priced ahead of itself. If China Telecom's involvement is purely financial with marketing-only co-construction language, the carrier-stack positioning was press-release inflation all along.

What to watch: the China Telecom vehicle named on the cap table; the equity-versus-order split in the term sheet; whether Ant Lingbo's commitment is exclusive; the first independent benchmark of MEgo's precision claim; and the cadence of the next round. Embodied-AI competition is moving from a model race to a data-infrastructure race. This round is the first one where the state-strategic layer is on the table.