In just two years, Qianxun Intelligence's valuation exceeded 10 billion yuan, but it still doesn't dare to relax.

The "Hundred-Billion Club" for humanoid robots has welcomed its sixth member.

On February 24th, Spirit AI announced that it had completed two consecutive rounds of financing totaling nearly 2 billion yuan, and its valuation exceeded 10 billion yuan.

Getting funding at this point in time is in itself a signal.

The list of investors is also thought-provoking: New shareholders such as Yunfeng Capital, Sequoia China, TCL Venture Capital, and 360 Fund have entered the scene, while old shareholders like Shunwei Capital, Prosperity7, and Fortune Capital have continued to increase their investments. Those familiar with the primary market know that this structure of "both new and old investors betting" means that institutions are no longer just onlookers but have clearly taken sides.

Before 2026, there were only three members in the "Hundred-Billion Club" for humanoid robots: Yuan Robotics founded by well - known Bilibili UP master Zhihui Jun, Unitree Technology which appeared on the Spring Festival Gala, and Galaxy Universal, standing in a tripartite confrontation. In 2026, Xinghaitu, Zhipingfang, and Spirit AI successively joined the "hundred - billion table" around the Spring Festival.

In other words, a valuation of 10 billion yuan is no longer an isolated case in the humanoid robot field but has begun to form a sector effect.

Han Fengtao, the founder, said quite straightforwardly in an interview with LatePost AI: "The embodied AI in 2026 will be very similar to the large models in 2023. If you can't get a lot of funding and your model performance can't reach the top, you won't have a chance to enter the game."

To put it another way: In the humanoid robot race, financing has changed from an 'accelerator' to a 'ticket'. Without sufficient funds, it means direct elimination.

Spirit AI getting 2 billion yuan at this time reflects a subtle market sentiment to some extent: Capital is still willing to bet on the future, but the premise is that you must already be in the first echelon.

A valuation of 10 billion yuan is not just a label.

It is an admission ticket and also a pressure gauge. The countdown starts the moment you get it.

The Crazy Record of Reaching a 10 - Billion Valuation in 2 Years

Among the "Hundred-Billion Club", Spirit AI, founded in 2024, is not only one of the youngest companies, but it only took 26 months from its establishment to reaching a valuation of over 10 billion yuan.

Its financing curve has hardly paused.

In January 2024, the company was established. It completed nearly 200 million yuan in seed and angel rounds of financing within 7 months. In 2025, it successively completed 528 million yuan in the Pre - A round and nearly 600 million yuan in the Pre - A+ round of financing. At the beginning of 2026, it received nearly 2 billion yuan. In six rounds over two years, the total financing was about 3.328 billion yuan.

This pace is more like "racing against time" than just "high - speed".

In contrast, Unitree Technology, founded in 2016, took 9 years to reach a valuation of over 10 billion yuan. Spirit AI compressed this cycle to just over two years. In essence, it's not an improvement in efficiency but a different environment: It is riding on the window period after the explosion of large models.

However, Spirit AI did not emerge out of nowhere.

Han Fengtao, the founder, is a serial entrepreneur in the robot industry. In 2015, he co - founded the collaborative robot company LuoShi and served as the CTO, leading the team to deliver dozens of models and a cumulative total of over 20,000 industrial robots.

According to him, in the past decade, he witnessed "the domesticization rate of industrial robots increase from 3% to over 50%, but most companies are not making money."

The core problem of not making money is not that there are not enough robots, but that their working ability is limited.

Industrial robots are highly dependent on customization and fixed scenarios. They need to rewrite the logic when changing tasks. Their generalization ability is weak, the scale effect is limited, and the gross profit is repeatedly diluted. Han Fengtao also tried the route of "hardware upgrade + early - stage deep learning" back then, but limited by the upper limit of AI capabilities at that time, he could never create a truly "general - purpose brain".

The real turning point occurred in 2023. When ChatGPT appeared and general models showed the ability to transfer across tasks, he realized that a variable had emerged.

"Reducers, integrated joints, motors, complete machines... but these are not the biggest opportunities for humanoid robots. The biggest opportunity is the brain, so I'll find out who can develop the brain."

In the summer of 2023, Han Fengtao met with more than 100 people and finally found Gao Yang, one of the "Four Returnees from Berkeley" and the director of the Visual and Humanoid Robot Laboratory at Tsinghua University. "Gao Yang has been researching end - to - end autonomous driving since 2017 and knows how to use Internet videos for training. He is absolutely reliable in terms of technology."

This was a route selection.

Investors once commented that the team configuration of Spirit AI is comparable to that of Pi and Google DeepMind. Whether this evaluation holds or not, the core is that it is betting on the integration path of "large models + embodied AI".

In January 2024, Spirit AI was established. In July of the same year, Moz0, the first - generation robot of Spirit AI, was born. In August, Spirit AI completed nearly 200 million yuan in seed and angel rounds of investment.

In March 2025, the preview version of the Spirit V1 VLA model was released. The humanoid robot equipped with this model can complete long - range complex tasks such as "folding clothes". As soon as the technological milestone appeared, the financing followed immediately, and the 528 - million - yuan Pre - A round was quickly finalized.

In January 2026, the company further open - sourced the Spirit v1.5 model. In the Table30 list of RoboChallenge, Spirit v1.5 scored 66.09 points with a success rate of 50.33%, surpassing Pi0.5 and becoming the first Chinese open - source embodied model to outperform Pi0.5 in a public benchmark test. Subsequently, nearly 2 billion yuan in financing was in place.

Looking back at Spirit AI's rapid financing, a clear rhythm can be seen: Every leap in model performance corresponds to an increase in valuation.

This is a typical "technology - driven financing curve". But it also means higher risks: Once the performance iteration slows down, the patience of capital will quickly decline.

The Big Bet on Data Collection

Unitree's performance of "WuBOT" on the Spring Festival Gala reignited the emotional value of humanoid robots; Spirit AI's reaching a 10 - billion valuation in two years quietly raised the ceiling of the entire industry. In 2026, there is no doubt about the popularity of humanoid robots.

However, Han Fengtao, who has obtained the financing, is not relaxed. In his view, this popularity is more like a watershed than a carnival: "The embodied AI in 2026 is very similar to the large models in 2023. If you can't get enough funds and your model performance can't reach the top, you won't have a chance to enter the game."

The subtext is clear: The window period is rapidly narrowing. With the double pressure of capital and technology, once you fall behind, the possibility of a comeback is almost negligible.

After getting the funding, Spirit AI has set a more aggressive goal for itself: To rank among the top 3 global embodied brains and increase the scale of effective training data to 1 million hours. The best open - source model before only used 10,000 hours, so this is a 100 - fold expansion.

Different from language models that rely on Internet texts, robot models need real - world interaction data from the physical world: repeated cycles of action execution, environmental feedback, force - control errors, and failure correction. Every grasping failure and every path deviation will become a training sample for the model. The scale and structure of the data directly determine the upper limit of generalization ability.

The problem is that this is a very costly path. Real - world data cannot be "crawled"; it requires hardware, venues, manpower, and time. Once the scale increases, the cost will increase exponentially.

For this reason, Spirit AI is betting 80% of its resources on the data system, which is essentially a strategic gamble. Han Fengtao's judgment is straightforward: The reason why humanoid robots have not exploded is not that the algorithms are not smart enough, but that there is not enough data. If he had sufficient early - stage funds, he would not hesitate to 'All in on data'.

But the key difference is not "whether to need data" but "how to obtain data".

Spirit AI did not take the "small and refined" route, nor did it pin its hopes on a small number of laboratory samples. They chose an industrial approach that prioritizes scale and keeps costs under control.

Relying on Gao Yang's experience in autonomous driving and Internet video training, the team did not build a complex world model frame by frame. Instead, they first used a large number of Internet videos for pre - training, using fewer parameters to gain a higher starting point and significantly reducing computing power costs. This is actually a typical large - model approach: First, achieve wide coverage, and then perform physical alignment.

Offline data is also collected in multiple parallel ways: remote operation, wearable device collection, and real - scene playback. The goal is not to pursue "perfect samples" but to reduce data costs. It is reported that compared with traditional methods, the cost has been reduced by about 90%. Only when the cost is reduced can large - scale expansion be supported.

What's more interesting is that the team does not deliberately clean up "dirty data". Grasping failures, chaotic movements, and non - standard operations are all retained. Their logic is that the real world is messy. If the data is too clean, the model cannot learn the ability to handle complex scenarios. Compared with highly screened laboratory samples, this noisy and diverse data is more conducive to the model forming cross - scenario transfer ability and even approaching zero - sample generalization.

This training logic has been reflected at the model level. In March 2025, the Spirit v1 model was released, which completed long - range flexible object manipulation - folding clothes is no longer a single grasping action but a complete process of continuous action planning and execution. The key is not "grasping it" but "continuously doing it right".

In December of the same year, Moz completed continuous tasks such as desktop tidying, garbage throwing, seat returning, and microwave heating in the rehabilitation and elderly - care track of the 2025 Intelligent Robot Competition. What was demonstrated was not a single - point skill but a cross - step decision - making chain.

Han Fengtao, who has come from the era of industrial robots, is clearer than most people that what really determines commercial value is never a dazzling display but the ability to repeat stably 1,000 or 10,000 times. Only the ability to be repeatable, transferable, and scalable can bring real commercial value.

"All in on data" is the most rational choice at the current stage, but it is also a high - risk bet.

Because once the scale is maximized but the performance does not increase linearly, the patience of capital will be quickly exhausted. The market can pay for technological visions, but it will not pay for efficiency problems indefinitely.

Hitching a Ride with CATL Is Not Enough

However, Spirit AI, which can fold clothes, has chosen to enter the factory first.

Although Gao Yang proposed the "Double - Ten Plan" in 2024, aiming to "enable 10% of the world's population to own their own robots in 10 years", in reality, Spirit AI did not bet on the household scenario first but returned to "working in the factory".

Behind this is a pragmatic calculation of the risk - return ratio.

The household scenario has great imagination space, but there are many variables, the willingness to pay is unclear, and the cost structure is not well - defined; the industrial scenario has clear tasks, clear standards, and quantifiable acceptance. For a company that has been established for less than two years, cash flow and verification speed are more important than the narrative.

Han Fengtao's view of industrial robots is not romantic. After ten years of in - depth involvement, he saw the reality of an "unprofitable" industry: Traditional industrial robots have single functions and can only complete fixed processes, and their penetration rate is increasing slowly. There are about 100 million manufacturing workers in China, but the number of industrial robots in use is only about 3 million, with about 300,000 new ones added each year. Structural substitution has never really happened.

His inference is that if the embodied brain can replace 30% of the manual labor, the scale of the automation market will experience an order - of - magnitude leap.

In December 2025, the Zhongzhou Base of CATL was officially put into operation, known as the world's first large - scale new - energy power battery production line for humanoid robots. Spirit AI's Moz robots are responsible for battery plug - in and detection tasks there.

It only took 22 months from the company's establishment to entering the industrial large - scale scenario. According to Han Fengtao, "It took us 11 months from the initial contact with CATL to the final acceptance. Spirit AI is the first embodied company that can really work on the battery production line, and so far, it has participated in the production of over 1,000 batteries."

Compared with demonstration videos, this "participation in production" is more valuable. Because the standard of the industrial system is not smooth movements but stability, yield rate, and rhythm matching. Being able to enter the production line means that the system is controllable within a certain range.

But this does not mean safety.

Among Spirit AI's investors, there are JD.com and 360 Fund, as well as Shunwei Capital and Yunfeng Capital, which are connected to the ecological systems of Xiaomi and Alibaba respectively. Han Fengtao often says within Spirit AI: "Our real future competitors are these big companies. Companies like Huawei, Xiaomi, and Li Auto, which do both hardware and software, are more likely to succeed."

Therefore, "reaching a 10 - billion valuation in two years" is not the end but a countdown. Han Fengtao's goal is straightforward: Before the big companies fully enter the market, make Spirit AI a'medium - sized company' and sell at least 100,000 robots a year. This is a quantifiable indicator for a medium - sized company, corresponding to economies of scale, cost dilution, and bargaining power.

The capital environment is also tightening. Although two more humanoid robot companies reached a 10 - billion valuation at the beginning of the year, the overall financing enthusiasm in the track has begun to decline. Public data shows that in the third quarter of 2025, the financing amount in the humanoid robot track decreased by 42% year - on - year, the number of financing events decreased by 35%, and the proportion of early - stage project financing dropped from 68% in 2024 to 32%. The market is shifting from "betting on concepts" to "looking at realizations".

This is also a sign that humanoid robots are moving from the concept stage to the verification stage.

After watching a Spring Festival Gala with a high "robot content", the 2 - billion - yuan financing obtained by Spirit AI is, to some extent, an epitome of this structural shift: The market still bets on the future, but it no longer pays for imagination but for the ability to "work in the factory".

When the industry moves from performances to production lines, the real test for humanoid robots is just beginning.

Reference materials:

LatePost AI, "Exclusive Interview with the Founder of Spirit AI: New 2 - Billion - Yuan Financing, Elimination Tournament of Embodied Models, and Non - Consensus on Implementation"

Deep Blue Finance, "A Blockbuster at the Beginning of the Year! It Ran the Road That Unitree Took 9 Years to Run in 2 Years. Another Unicorn Emerges in Hangzhou"

Pencil News, "An 80s - Born from