Hundreds of billions of unicorns are emerging in a blowout, and embodied intelligence is shifting towards the "brain".

Just as the hustle and bustle of the Spring Festival in the Year of the Horse subsided, the embodied intelligence track kicked off a new round of competition in 2026 with a financing feast led by “national teams.”

On March 2nd, two financing announcements set off a storm in the industry on the same day.

First, Galaxy General Robotics, a star enterprise on the Spring Festival Gala stage, announced the completion of a new round of financing worth 2.5 billion yuan. The list of investors was top - notch. The National Artificial Intelligence Industry Fund (the third phase of the National Big Fund) made its first investment in the embodied intelligence track, and super state - owned enterprises such as Sinopec, CITIC Group, and Bank of China were among the investors.

Immediately afterwards, Songyan Power, also on the Spring Festival Gala stage, announced the completion of a nearly 1 - billion - yuan Series B financing, led by Chendao Capital, an industrial platform affiliated with CATL.

This is not an isolated case. If we look back to around the Spring Festival, companies like Qianxun Intelligence, Independent Variable Robotics, Zhifang, and Xinghaitu successively announced financings exceeding one billion yuan.

According to incomplete statistics, since the beginning of 2026, the total disclosed financing in the field of embodied intelligence has approached 15 billion yuan, and the number of domestic embodied intelligence companies with valuations exceeding 10 billion yuan has increased to seven: Unitree Technology, Zhipu Robotics, Galaxy General, Xinghaitu, Zhifang, Qianxun Intelligence, and Independent Variable Robotics.

On one hand, questions about “over - inflated bubbles” are constantly being raised. On the other hand, capital has created seven peaks of 10 - billion - yuan valuations with real money. More intriguingly, on February 28th, China's first national - level “Standard System for Humanoid Robots and Embodied Intelligence (2026 Edition)” was officially released, marking that the industry has entered a new stage of standardization from wild growth.

This feast that started in 2024 and fully erupted in 2025 not only shows no sign of cooling down at the beginning of 2026 but is intensifying. Behind the noisy numbers, a more profound change is taking place: the spotlight of capital is quietly shifting from the “physical” dazzling skills to the hidden revolution of the “brain.”

01

From “Limbs” to “Brain”

The Technological Shift Behind the Billion - Yuan Financing

It was Zhifang, headquartered in Shenzhen, that first set off a wave in the market.

On February 23rd, the company announced the completion of a Series B financing exceeding 1 billion yuan. The list of investors included Baidu Strategic Investment, CRRC Capital, etc.

This company, which has been established for only three years, focuses on its self - developed “Global Whole - body VLA Large Model” GOVLA. Different from the past practice of simply splicing visual, language, and action modules, the core of GOVLA lies in its end - to - end multi - modal fusion architecture. It enables robots to handle the three tasks of “what to see,” “what the instruction is,” and “how to coordinate body movements” simultaneously, thus achieving a leap in the generalization ability for cross - scenario tasks.

The AlphaBot series of robots equipped with this model have now entered the industrial flexible manufacturing scenario.

Next up is Qianxun Intelligence.

This Hangzhou - based company has completed two consecutive rounds of financing totaling nearly 2 billion yuan. The investors include Yunfeng Capital, Sequoia China, etc. Its technological breakthrough lies in the open - sourcing of its self - developed Spirit v1.5 model. Based on VLA, this model strengthens the fine - grained modeling of physical interactions, enabling robots to handle more complex dynamic tasks.

A key verification is that at the end of 2025, its humanoid robot “Xiaomo” was put into operation on the world's first humanoid embodied intelligence production line at CATL's Zhongzhou Base, achieving zero - fault plugging and detection of nearly a thousand batteries. This is behind the model's breakthrough in high - precision force control and real - time environmental adaptation capabilities.

The industry changes revealed by these two financings are also evident in Xinghaitu, which completed a 1 - billion - yuan financing before the Spring Festival.

This company, founded by a team with backgrounds from Waymo and Momenta, launched an upgraded version of the end - to - end VLA basic model, G0Plus, in January this year. Its technical route emphasizes “action” as the core, trying to break through the limitations of language models and enabling robots to directly learn causal logic through actions in the physical world.

Independent Variable Robotics, which has the highest valuation among the companies that have announced financing in 2026, has taken a different path. The company has raised more than 3 billion yuan in total in more than two years since its establishment. Its launched native multi - modal input - output architecture, WALL - A, attempts to break through the “last nanometer” barrier between vision, language, and the robot's underlying control instructions, enabling AI understanding to be more directly translated into the robot's precise actions.

Zhang Yu, a partner at Qingzhi Capital, recalled that when investing in Zhifang in 2023, the “embodied brain” route was considered “too far - ahead,” and it even experienced a financing drought for more than a year. The turning point came in the second half of 2025. With the maturity of the VLA (Vision - Language - Action) multi - modal technology path, the success rate of robots in cross - scenario tasks increased significantly, and the technical certainty of brain - focused companies was finally recognized by the capital market.

So far, almost all of the newly added four billion - yuan unicorns are pointing in the direction of the “embodied brain,” and the focus of capital is shifting from the flexibility of “limbs” to the intelligence of the “brain.”

02

Why Can't the Capital Fever Stop?

In the current context of a tightening macro - financing environment, why can the embodied intelligence track still attract capital against the trend, with single - deal amounts often reaching billions of yuan? Behind this is the resonance of three logics: technology, industry, and perception.

First, the convergence and breakthrough of technical routes have brought “certainty dividends.”

If in 2024 the industry was still debating “whether humanoid robots are necessary,” then after 2025, a consensus is gathering around the VLA multi - modal large model. Whether it is Google's RT - 2 or various models self - developed by domestic enterprises, the end - to - end VLA architecture has been proven to be the key path for robots to break free from “preset programs” and gain generalization ability.

Galaxy General Robotics, which announced a new round of 2.5 - billion - yuan financing on March 2nd, is a typical example of a company that has received heavy capital investment due to a breakthrough in its technical route.

This company, founded by Wang He, a researcher at Peking University, has chosen a different path from the industry mainstream: a virtual - real fusion training paradigm that mainly uses synthetic simulation data and supplemented by real - machine data. Its self - developed “Galaxy Star Brain” (AstraBrain) is the world's first full - body and full - hand end - to - end embodied large model that integrates “brain - cerebellum - neural control” into one model, opening up the entire link from high - level multi - modal perception to low - level real - time feedback control.

Based on the world's largest embodied intelligence dataset with tens of billions of samples, Galaxy General has achieved the generalization ability for humanoid robots to “draw inferences about other cases from one instance” in new scenarios and with new objects with only a small number of samples or even zero samples. The success rate of the model is as high as 99%, far exceeding that of global peers.

It is precisely this technological breakthrough that reconstructs the physical intelligence paradigm from the underlying architecture that has attracted the National Artificial Intelligence Industry Fund (the third phase of the National Big Fund) to make its first investment in the embodied intelligence track, and industrial capitals such as Sinopec, CITIC Group, and SAIC Financial Holdings have also entered the scene.

Zhang Zhikuan, a director of the investment department at Cornerstone Capital, pointed out that as the success rate of robots in cross - scenario tasks such as object grasping and pouring coffee has increased significantly, “brain - focused companies are beginning to show stronger technical certainty.” This leap from “scientific research exploration” to “engineering verification” has given industrial capitals that dare to make large - scale investments a sense of security.

The prosperity and competition of technical routes are in themselves a proof of the industry's vitality.

Second, the prototype of a commercial closed - loop has been proven in real - world scenarios.

The capital market has always been pragmatic. The entry of industrial capitals such as CRRC Capital, SAIC Financial Holdings, and CATL is not simply chasing concepts. In addition to Qianxun Intelligence's implementation on CATL's production line, where robots replace humans in battery plugging and detection, with a rhythm comparable to that of skilled workers and zero - fault mass production; Zhifang has reached a cooperation with the Huike system, planning to deploy more than a thousand robots in three years, with an order value approaching 500 million yuan; Xinghaitu claims to have achieved thousands of orders.

These numbers mean that embodied intelligence has stepped out of the laboratory and begun to create real value in scenarios such as industrial flexible manufacturing and warehousing logistics. For manufacturing giants pursuing cost reduction and efficiency improvement, investing in a robot company is essentially investing in future productivity.

Third, the “popular science effect” of the Spring Festival Gala and the “base support” of the supply chain have had a double - impact.

On the 2026 CCTV Spring Festival Gala, robots from four companies, Unitree, Songyan, Magic Atom, and Galaxy General, competed on the same stage. From backflips to interacting with humans, they completed a national popular science campaign covering 1.4 billion people. This not only improved public awareness but also sent a signal to investors: China's humanoid robot industry chain is mature enough to support the concentrated appearance of multiple brands and scenarios.

It is generally recognized in the industry that humanoid robots and smart cars have a 70% overlap in the supply chain in areas such as batteries, motors, sensors, and die - castings. As a major new - energy vehicle hub in China, Chongqing has attracted the southwest headquarters of Unitree, the innovation center of Galaxy General, and the factory of Zhipu to cluster there, forming a unique “automobile - to - robot - making” ecosystem. As the localization rate of core components increases and costs continue to decline, the influx of capital is a natural result.

03

Behind the Hustle and Bustle, It's Not Time to Decide the Winner

However, the excitement belongs to the capital. For entrepreneurs, a billion - yuan valuation is not only a medal but also a sword hanging over their heads.

The first question is the “end - game dispute” over technical routes.

Although VLA is currently in the limelight, the doubts have never disappeared. Wang Xingxing, the founder of Unitree Technology, once described the common VLA models in the industry as “foolish architectures.” The focus of the controversy is: Is language really the best intermediary for robots to understand the physical world?

Zhao Xing, the CTO of Xinghaitu, believes that robots need a Large Action Model centered around “action” rather than simply relying on language.

Meanwhile, a new paradigm represented by the “world model” is emerging. It allows robots to pre - simulate physical evolution in an internal imaginary space and then output actions. Nvidia's Cosmos and LingBot - VA open - sourced by Ant Lingbo are both exploring this “predict - then - act” path.

Zhao Mingguo, a professor in the Department of Automation at Tsinghua University, warned that the VLA highly favored by the industry may only be a transitional means. The visual and tactile data in the physical world are far from being as standardized as language data, and the experience of large language models cannot be simply copied.

The more fundamental challenge is that the visual and tactile data in the physical world are far from being as standardized and abundant as language data. The scarcity of high - quality physical interaction data remains a common pain point in the industry. Compared with the tens of millions of data required for autonomous driving, the data volume required for embodied intelligence is more than ten times that, and the cost of data collection is extremely high.

The second question is the “scale gap” in the business model.

As of now, there may be less than 1,000 robots truly operating in human work scenarios globally. Jiang Lei, the chief scientist of the National - Local Joint Humanoid Robot Innovation Center, admitted that the industry currently does not dare to produce more than 10,000 units per year because “producing too many has no use, and the after - sales pressure will be very high.”

Compared with the tens of millions of data required for autonomous driving, the data volume required for embodied intelligence is more than ten times that, and the scarcity of high - quality physical interaction data remains a common pain point in the industry.

A more realistic question is: How long of an investment return period is needed to convince enterprise customers to buy a robot that costs hundreds of thousands or even millions of yuan? When capital shifts from being “story - driven” to “financial - report - driven,” the answer to this question will determine who can still stay in the game.

The third question is the “irrational exuberance” in the valuation logic.

Independent Variable Robotics has raised more than 3 billion yuan in total in more than two years since its establishment. Xinghaitu has completed eight rounds of financing since its founding, with a group of star institutions such as Ant, IDG, Hillhouse, and Mihayo standing behind it. These high valuations correspond to extremely optimistic expectations for the future rather than the current profit reality.

As Zhang Zhikuan of Cornerstone Capital said, “It's still far from the time to decide the winner.” For these startups, most of the financing funds will continue to be invested in model training and R & D. Although commercialization is important, the model's ability determines the upper limit. When the industry is still burning money for the evolution of the “brain,” any misjudgment of the technical route or missed opportunity in commercialization may cause the billion - yuan valuation to collapse instantly.

04

Conclusion

The emergence of two billion - yuan unicorns in two days is not only a “good start” at the beginning of the Year of the Horse but also a microcosm of the embodied intelligence industry moving from concept to actual combat. As the tide of capital shifts from the “physical body” to the “brain,” and from “show - off skills” to “implementation,” the industry is undergoing a silent and profound transformation.

The future competition will no longer focus solely on the financing speed or valuation figures. Instead, it will be about who can be the first to cross the “reliability gap” and run a replicable commercial closed - loop in real - world scenarios; who can find the optimal solution between data cost and algorithm iteration; who can bet on the real path to AGI in the paradigm dispute between VLA and the world model.

For entrepreneurs in this field, a billion - yuan valuation is just the beginning. The real test lies in how to transform these numbers into stable - running screws in the physical world, cups of coffee that won't spill, and contracts that customers are willing to renew.

After all, when the stage lights go out, the roar in the workshop is the only real sound.

This article is from the WeChat official account “Insight New Research Club” (ID: DJXYS - 0309), written by Chen Wen, and is published by 36Kr with authorization.