Why Did the “National Team” Collectively Bet $300 Million on This Robotics Startup?

On March 10, Lingchu Intelligent announced the completion of angel and Pre-A rounds totaling 2 billion yuan.

Even within the highly capitalized embodied intelligence sector, this figure is striking. Yet more noteworthy than its scale is the lineup of investors behind it: the angel round simultaneously attracted “national team” capital, including China Development Bank Financial Assets Management Co., Ltd., China Zhongtou Capital, and the CCTV Integrated Media Industry Investment Fund; industry leaders and renowned investment funds followed closely. The Pre-A round further secured backing from state-owned capital in multiple regions, including Shanghai and Wuxi.

Why has a company established for just over a year attracted the participation of “national team” investors and industrial capital consortia? The underlying logic behind this may be more worthy of analysis than the $200 million figure itself.

Who is Lingchu Intelligence?？

Lingchu Intelligent was established in 2024, and its core founding team gathers top talents in the industry.

Founder and CEO Wang Qibin has extensive experience in the robotics and consumer electronics industries, having previously served as President of JD.com’s Robotics Division and Vice President of Products at CloudMinds Technology. He has also worked for companies including ForwardX Robotics, BlackBerry, and Sonos, accumulating over 20 years of hands-on experience in hardware development and commercialization.

Co-founder Chen Yuanpei, known as a "00s generation tech genius," is a top student of Professor Fei-Fei Li, the "AI goddess," and has passed the selection for Huawei's "Talent Youth" program. He has achieved several breakthrough results in the fields of reinforcement learning and dexterous robot manipulation.

The other co-founder,, with over 15 years of experience in the robotics and autonomous driving fields, has held core technical positions at leading tech companies such as Tencent, Alibaba, and JD.com. He is proficient in algorithms, simulation, engineering, and full-stack technology. Note: The name "" is kept in its original form as it is a proper noun. If you prefer, it can be transliterated as "Chai Xiaojie", but this requires confirmation from the person or official documents.

In addition, Lingchu Intelligent has hired Yang Yaodong, assistant professor at the Peking University Institute of Artificial Intelligence, as Chief Scientist.

It can be seen that this is a "dream team" with deep technical expertise and rich practical experience in the fields of robotics, artificial intelligence, and commercialization.

In terms of strategic positioning, Lingchu Intelligence has chosen a path distinct from most peers: it does not develop full-stack hardware platforms, but instead focuses on a “compact full-stack” approach.

Known as "small full stack," it focuses on building a software and toolchain system centered around end-to-end VLA models, dedicated to combining dual arms, five-finger dexterous hands, and wheeled platforms, aiming to solve the dexterity operation challenges in embodied intelligence.

In terms of scenario selection, Lingchu Intelligent focuses on semi-structured logistics and retail scenarios.

Subsequently, Lingchu Intelligence’s logic is as follows: The home scenario involves too many extreme cases to achieve short-term (closed-loop), while the data value in traditional factory scenarios remains relatively isolated. In contrast, the logistics scenario features both high-frequency, generalizable demands and numerous non-standard tasks requiring dexterous human operations—making it an ideal entry point for embodied intelligence to demonstrate its value.

Based on disclosed information, Lingchu Intelligence has completed small-scale scenario validation at real logistics customers’ warehouses, achieving improved sorting efficiency.

This means that its technical approach has been preliminarily validated in real-world environments and has entered the transition phase from the laboratory to production lines.

Attracting intensive capital attention？

A keyword is "data".

In the embodied intelligence industry, a consensus is emerging: high-quality real-world data is the key bottleneck on the path to general intelligence. Even Lingchu Intelligence views the current competition in embodied intelligence as, at its core, a competition for high-quality data.

In terms of data acquisition, there are currently several mainstream approaches in the industry: one is to generate a large amount of simulated data through simulation. Although this method is low-cost and can quickly obtain large-scale data, simulated data cannot fully replace real-world physical interactions, especially real-world physical feedback such as force, friction, and changes in center of gravity, which cannot be replaced by simulation.

Another approach is to collect and annotate data manually. This method can ensure the authenticity and accuracy of the data, but it requires a large amount of human, material, and time resources. Moreover, it is difficult to cover complex and changing real-world scenarios, and the diversity and generalization ability of the data are limited.

Another approach involves leveraging the vast amount of video resources available online, extracting visual information from these videos to train models. This method has relatively low acquisition costs and provides exposure to a rich variety of scenes and actions. However, critical elements such as an object's texture, weight, and the sense of force during manipulation remain missing, often leading to situations where the model "understands by watching but fails in actual execution."

Addressing these pain points, Lingchu Intelligence has chosen to focus on "real human data collection." In Lingchu Intelligence's view, the human execution structure dictates that large-scale data must inevitably originate from everyday human operations.

To fully collect this data, Lingchu Intelligence has independently developed the embodiment-native human data collection solution Psi-SynEngine, which comprises a portable exoskeletal tactile glove data acquisition kit, a large-scale “in-the-wild” data collection pipeline, and a cross-embodiment data transfer model based on world models and reinforcement learning.

The exoskeleton tactile glove, independently developed by Lingchu Intelligence specifically for data collection, can achieve a positioning accuracy of up to sub-millimeter level. It is capable of fully capturing all degrees of freedom of the hand and arm, as well as tactile information covering the entire hand, without affecting the normal operation of the personnel.

In order to support this data acquisition solution, Lingchu Intelligence has also built a pipeline and platform capable of processing large-scale data, cooperating with its self-developed large model to achieve high-precision data annotation and post-processing, forming a complete data production cycle.

Source: Lingchu Intelligence

According to Lingchu Intelligence, the Psi-SynEngine can directly collect operational data from frontline workers in real-world working environments—including logistics, factories, supermarkets, hotels, and households—without requiring secondary data migration. This means that, compared to traditional data collection solutions, this approach offers superior portability, higher data collection efficiency, and lower costs.

According to Wang Qibin, CEO of Lingchu Intelligence, the integrated cost of data collection via gloves can be reduced to about one-tenth that of a real-machine teleoperation solution. In the future, Lingchu Intelligence also plans to launch a portable, crowdsourced version, which is expected to further lower this cost.

Lingchu Intelligent has planned to build the largest dexterous hand dataset in the country this year. The two core investment areas of this 2 billion yuan funding are the large-scale application in logistics scenarios and the construction of a large-scale data acquisition solution system.

Notably, even with sufficient data collected, bridging the inherent structural and capability differences between human hands and robotic hands remains a significant challenge, and this is precisely where Lingchu Intelligent's Psi-SynEngine creates its competitive advantage.

It is reported that by 2025, relying on Psi-SynEngine, Lingchu Intelligence has rapidly constructed a Psi-SynNet-v0 dataset of tens of thousands of hours scale within the company, and plans to further break through the one million hours scale this year, aiming to become the world's largest dexterous operation dataset.

Meanwhile, this year Lingchu Intelligence also plans to further advance the implementation of embodied intelligence in complex logistics environments and promote process-oriented delivery, enabling humanoid robots to truly enter factories.

In the embodied intelligence field, whoever can obtain more and higher quality human data at a lower cost will be better positioned to establish the iteration flywheel from "data" to "model" and then to "scenarios".

From the disclosed information, Lingchu Intelligence's logic is being preliminarily validated in logistics scenarios.

But the $2 billion funding is just the beginning; the real test lies in whether this flywheel can keep spinning and extend beyond sorting efficiency gains into more complex industrial scenarios.