3.5 billion in three months: Investors are scrambling to back the "OpenAI of the physical world"

A familiar scene is playing out once again.

The investment community has learned that Giga Vision has announced the completion of another 1 billion yuan in Series B2 financing. This round of financing was jointly invested by top global national team funds, industrial capitals, financial institutions, and state-owned platforms such as Lion City Capital (a top cross - border investment institution from Singapore that has continuously invested in multiple rounds), China - Belgium Fund (Sino - Belgian Fund), CIC Investment, Wanxiang Qianchao, Fosun RZ Capital, Huagai Capital, Jinchuangtou, Deyi Capital, Huacang Capital, Yuanshi Fund, etc. Multiple old shareholders such as Guozhong Capital, Fortune Capital, and Turing Asset Management continued to increase their investment significantly.

According to insiders, the investment intention in this round of the market far exceeded the original financing target. It is worth mentioning that this is already Giga Vision's third round of financing within three months, with a cumulative amount of up to 3.5 billion yuan.

So far, Giga Vision and its 90 - post doctor Huang Guan, the helmsman behind it, have created a hot scene in the venture capital circle this year. Behind the collective bet of investors, perhaps the "GPT - 3 moment" of physical AGI is approaching.

Investors are lining up, 3.5 billion yuan in financing in just three months

As you can see, almost all types of top - tier investment institutions in the market have appeared behind Giga Vision.

As early as its establishment, Giga Vision received tens of millions of yuan in seed - round financing from Chentao Capital. Since then, investors have started to line up. In September 2024, it completed nearly 50 million yuan in consecutive angel and angel + rounds of financing, invested by institutions such as BAIC Capital, Miracle Plus, Huamin Investment, Longding Investment, Qingzhi Capital, and PKSHA Algorithm Fund.

After a year, in August 2025, Giga Vision received several hundred million yuan in consecutive Pre - A and Pre - A+ rounds of financing. The Pre - A round was led by Guozhong Capital, with Zifeng Capital and old shareholder PKSHA Algorithm Fund following. The Pre - A+ round was invested by CICC Capital, Guangzhou Industrial Investment, Yicun Songling, and Huaqiang Capital.

Subsequently, Giga Vision's financing rhythm became more intensive. In November of the same year, Giga Vision completed a new round of 100 - million - yuan - level Series A1 financing, jointly invested by Huawei Hubble and Huakong Fund. One month later, the company completed 200 million yuan in Series A2 financing, led by Fortune Capital and jointly led by old shareholder Huakong Fund. Well - known institutions such as Shoufa Development Venture Capital, Puyao Xinyie, Caixin Capital, Huajin Capital, Zhangke Yaokun, and Fuzhuo Venture Capital followed, and old shareholder Hedinggong Capital increased its investment significantly.

In 2026, Giga Vision's financing rhythm left a deep impression on the venture capital circle.

First, in early March this year, it completed nearly 1 billion yuan in Pre - B round financing. The investors included top chip and automotive industry capitals such as SMIC Juyuan, Shanghai Semiconductor Industry Investment Fund, Linxin Capital, Xingyuan Capital, and Wanlin International, as well as important state - owned platforms and well - known financial institutions such as CICC Capital, Suzhou Venture Capital, Huaqiang Capital, Yangtze River Capital, Optics Valley Industrial Investment, Xishan State - owned Investment, Jinyu Maowu, Xinding Capital, Lingyang Investment, Caixin Capital, Zhangke Yaokun, and Chengzhu Investment. Among them, CICC Capital, Huaqiang Capital, Caixin Capital, Zhangke Yaokun and other old shareholders continued to increase their investment significantly.

Immediately afterwards, in April, Giga Vision's Series B1 financing came to light. It was jointly invested by a well - known technology giant, multiple top national team funds, Jianling Capital (CVC of Yili Group), Puhua Capital, Huafu Investment, Yida Capital, New Industrialization Fund, Shengjing Jiacheng, Turing Asset Management, Kaiyang Capital, Wuhan High - tech Investment, Guiyang Jintou, Shandong Industrial Investment and other top - tier state - owned platforms, industrial capitals, and dual - currency financial institutions. Multiple old shareholders such as Huakong Fund, Huamin Investment, Yicun Capital, and Lingyang Investment continued to increase their investment significantly.

So far, Giga Vision's valuation has exceeded 10 billion yuan, making it the first domestic world - model unicorn worth over 10 billion yuan.

Until this time, the Series B2 financing was officially unveiled, which means that within just three months, Giga Vision has cumulatively completed 3.5 billion yuan in financing, and investors have voted with real money.

Looking back carefully, the financing process of Giga Vision since its establishment is just a microcosm of the continuous warming of the primary market's confidence in the physical AGI track, and it is also a firm vote of confidence from investors in Giga Vision's "world - model - driven physical AGI technology route + productivity - level implementation ability".

The signal revealed therein is also of far - reaching significance. Further analysis shows that this is not only a testament to Giga Vision's past technological accumulation but also the most valuable endorsement for it to lead the physical AGI track and develop a new industrial pattern. It is foreseeable that more investors will gather behind Giga Vision in the future.

The "dual - pyramid" system: the confidence to move towards physical AGI

As the outside world wonders, why is it Giga Vision?

Investment is about investing in people. The helmsman behind Giga Vision is a 90 - post Tsinghua University doctor, Huang Guan. He graduated from Huazhong University of Science and Technology with a bachelor's degree, then entered the Institute of Automation of the Chinese Academy of Sciences to pursue a master's degree, and later became a doctor in the Department of Automation at Tsinghua University. In addition, he has worked at Horizon Robotics and Jianzhi Robotics, and has also worked at institutions such as Microsoft Research Asia and Samsung China Research Institute.

What's even rarer is that in his past career, Huang Guan has cumulatively led or participated in financing of over 2 billion yuan. Therefore, Huang Guan is a rare composite leading talent in the industry who has top - level scientific research experience in the field of physical AI, mass - production engineering experience, business implementation experience, and continuous entrepreneurship experience.

The core team led by Huang Guan has also fully experienced the development process of physical AI in the past decade and has continuously achieved excellent results in technological innovation and industrial implementation at each stage, such as CV, autonomous driving, embodied intelligence, and world models. This is a rare "hexagonal warrior" team in the industry that has top - level experience and capabilities in aspects such as algorithms, data, ontology, mass production, business, and organization of physical AGI, and can be regarded as the "dream team" of physical AGI.

If talent is the booster for Giga Vision's rise, then technological innovation is the core foundation for it to stand in the global physical AGI track.

As is well - known, the R & D of physical AGI faces two core bottlenecks: one is data fragmentation, lacking high - quality, multi - dimensional data suitable for physical interaction scenarios; the other is that the language - dominated basic model is not an effective architecture for encoding 3D information, physical causality, and actions, making it difficult for the model to understand complex physical laws.

How to solve these two problems? The answer given by Giga Vision is to build a "dual - pyramid" system of algorithms and data with the world model as the core.

Among them, the data pyramid is divided into five layers, from bottom to top: Internet video data, real - person data, world - model simulator, simulated synthetic data, and real - machine data. This five - layer data architecture can solve the pain points of insufficient data, low quality, and single scenarios in physical AGI R & D, and provide sufficient and high - quality "fuel" for algorithm model training.

The algorithm pyramid is divided into three layers, mainly focusing on three core capabilities: world simulation, action alignment, and experience reinforcement. In this way, it can achieve the leap from physical cognition to entity execution and from passive execution to active evolution, enabling physical AGI to have learning and adaptation abilities similar to humans.

The core value of the "dual - pyramid" system lies in building a closed - loop evolution mechanism where data drives algorithms and algorithms feed back data. The data pyramid provides a large amount of high - quality physical interaction data for the algorithm pyramid, supporting the training and optimization of the algorithm model. The iterative upgrade of the algorithm pyramid can improve the accuracy of data collection and the authenticity of simulated data, and in turn enrich the content of the data pyramid.

More importantly, after three years of precipitation, Giga Vision has created a "world generation - action" dual - model system. Among them, the world action model transforms the understanding and prediction of the world model into the action strategy of the robot - GigaBrain - 0: a self - developed embodied VLA large model driven by the world model, which won the global championship in the world's largest real - machine evaluation RoboChallenge with a task success rate of 51.67%;

GigaBrain - 0.5M*: the world's first physical intelligent agent native paradigm centered on "world - model - dominated experience learning", which achieves self - evolution through "world model + reinforcement learning", with a success rate of nearly 100% in high - difficulty long - term tasks;

GigaWorld - Policy: a world action model that breaks the "speed - performance - efficiency" impossible triangle, achieving a 10 - fold increase in inference speed and training efficiency, and a nearly 30 - percentage - point increase in task success rate. It defeated Nvidia GR00T N1.5, PI0.5, etc. on the globally authoritative evaluation platform RoboCasa365 for home mobile operation tasks and won the global first place, becoming the first world action model to top the list.

The world generation model understands, simulates, and generates the physical world, providing data, a simulation base, and pre - training parameters for the action model - GigaWorld - 0: the world's first milestone work to verify that "world - model - generated data can effectively improve the performance of real robots", released and open - sourced in December 2025, with the GitHub open - source code getting more than 1.5k stars;

GigaWorld - 1: an action - conditioned world model (AC - WM), which won the global championship on the authoritative evaluation WorldArena with a comprehensive score of 62.34, defeating models from international top - tier institutions such as Google, Nvidia, and Alibaba, and is the first model on the list to break through 60 points;

DriveDreamer: the world's first autonomous driving world model for the real physical world, invited to give an oral presentation at NVIDIA and one of the most influential papers at ECCV 2024, and was the first to achieve large - scale industrial implementation of the world model.

Undoubtedly, the world action model and the world generation model are indispensable and are in a state of mutual complementation and spiral improvement, jointly forming the basic model of physical AGI, thus accelerating physical AGI towards the "GPT - 3 moment". To some extent, Giga Vision has embarked on a new path that is gradually being verified.

The physical world: the next stop for AGI

A new watershed moment in the AI era has arrived.

In the past few years, digital AGI has focused on information processing and virtual interaction, relying on large language models and multi - modal generation models to achieve functions such as text creation, image design, and code writing. In essence, it is an optimization and improvement of "information productivity".

The limitations are also obvious. Although digital AGI has greatly improved the efficiency of information dissemination, content creation, and data processing, it has always been unable to break through the boundary between the virtual and the real. As Fei - Fei Li, the "Godmother of AI", said, the large language model is still a "wordsman in the dark", eloquent but lacking experience, knowledgeable but not well - grounded.

Therefore, in the view of Giga Vision's team, AGI should not be confined to the screen. The core value of physical AGI lies in entity execution and physical transformation. It understands physical laws through the world model, perceives the physical environment through multi - modal means, and executes physical actions through mechanical bodies.

There is no doubt that GPT - 3 is widely recognized as the key node in the process of digital AGI implementation when the Scaling Law first showed emergent abilities. Today, after three years of continuous breakthroughs in the algorithm and data systems, Giga Vision has seen the trend of the convergence of the physical AGI route, which means that the "GPT - 3 moment" of physical AGI may soon arrive.

It is reported that Giga Vision's GigaBrain - 1 will be released in the third quarter of this year. As the world's first physical AGI basic model built based on the "dual - pyramid" system, GigaBrain - 1 will bring three key breakthroughs: visual native understanding (using vision as the main channel for state understanding), high - level language planning (language is responsible for high - level task decomposition), and physical law alignment (systematically expanding all types of large - scale training data).

After that, GigaBrain - 2 and GigaBrain - 3 will also be launched successively. Among them, GigaBrain - 3 will be trained based on 10 million hours of video data + 1 million hours of world - action data, aiming at the "GPT - 3 moment" of physical AGI.

Of course, technology ultimately needs to return to the realization of industrial value.

Giga Vision has found a unique way: entering households on the C - side and factories on the B - side, running on two tracks simultaneously. Looking at the industry, currently, there are very few embodied intelligence enterprises that can obtain household orders because the demands of real household scenarios are more complex and diverse, far less standardized than industrial scenarios.

However, Giga Vision still rises to the challenge. Not long ago, it launched the sub - brand "SeeLight" for household scenarios and introduced the first general humanoid robot "SeeLight S1" to enter real households, which has received orders for 100 units in real household scenarios and will be deployed in the Optics Valley Apartment Community in Wuhan first, starting large - scale operation in the third quarter. The next - generation household general robot "SeeLight S2" will also be released in the third quarter.

In this way, Giga Vision has achieved a breakthrough in the most scarce real - machine household data in the industry. Along the product rhythm of SeeLight S2/S3, it corresponds to the ChatGPT moment of physical AGI - making common skills widely used in real household scenarios.

On the B - side, on the one hand, for industrial manufacturing scenarios, Giga Vision is moving from single - point verification to large - scale mass production. In April this year, Giga Vision launched the full - stack self - developed physical AGI native general robot Maker H01 and, together with FAW Mold and Alibaba Cloud, completed the implementation of a full - process solution for embodied intelligent robots in real industrial manufacturing scenarios, reducing the scenario adaptation cycle of traditional automation solutions from several months to several weeks.

At the same time, Giga Vision announced this month that it plans to jointly deploy 1,000 general robots equipped with Giga Vision's world - model - embodied brain and Maker series in Wuxi with Longsheng Technology within three years. This is the world's first large - scale implementation of general robots driven by a physical intelligent basic model in industrial scenarios at the scale of 1,000 units, marking that domestic embodied intelligence has completely bid farewell to small - scale pilot projects and fully entered the large - scale mass - production cycle in industrial scenarios.

On the other hand, Giga Vision has long made the DriveDreamer series of autonomous driving world models a representative work in the industry. The new - generation driving simulator centered on the world model has reached signing agreements and mass - production cooperation with several domestic leading automakers, overseas and joint - venture automakers, as well as AI chip and Tier 1 giants, serving more than 30 leading automakers and autonomous driving companies at home and abroad.

In summary, the B - side layout represented by industrial series products corresponds to the Claude Code moment of physical AGI - the breakthrough of advanced skills