Gasgoo Munich- The embodied AI sector turned in a somewhat contradictory report card this past May.
On one hand, RoboScience secured 1 billion yuan in a Series A round, while early-stage players like Lumos Robotics, Vbot, and Five Ages also pulled in hundreds of millions. Big capital is still concentrating at the top. On the other, the total number of funding deals—and of those massive 1-billion-yuan-plus rounds—dropped sharply in May compared to previous months.
Behind these figures lies a brutal pivot for the entire industry: shifting from "selling a story" to "getting the job done."
May's funding list acts like a prism, refracting the tug-of-war between capital logic and industrial reality.
Capital "Cools," but the 'Water Sellers' Are Quietly Cashing In
According to incomplete data from Gasgoo, 22 funding rounds were disclosed in China's embodied robotics and core components sector in May. By comparison, March and April saw 35 and 31 rounds respectively—a clear downward trend.
Yet this cooling in May is largely in sync with sentiment across the broader primary market. In May 2026, there were 339 investment events in China's primary market—153 fewer than the previous month—while total investment capital fell 28.1% to 89.72 billion yuan.
In other words, the entire primary market hit the brakes in May.
Specifically within the embodied AI track, the largest round in May went to RoboScience, which secured 1 billion yuan at the Series A stage.
Founded in late 2024, RoboScience is dedicated to building a general-purpose embodied large model and has independently developed the VLOA model. Building on that, the company developed its own robot hardware to serve as a physical vessel for deploying the VLOA model at scale in the real world.
To some extent, RoboScience's hefty fundraise reflects capital's endorsement of its technological roadmap and deployment potential amid the ongoing investment boom in "embodied brains."
Notably, this marks RoboScience's second funding round this year, following a several-hundred-million-yuan Pre-A round in February. Since 2025, the company has closed a total of four rounds—a clear testament to investor confidence.
Another 1-billion-yuan round went to TETROBOTICS.
Image Source: TETROBOTICS
However, TETROBOTICS has reached the Series B stage. Given that 1-billion-yuan Series A rounds are becoming commonplace, this size isn't unexpected. Moreover, that 1 billion yuan figure represents the combined total of its Series B and B+ rounds.
Additionally, a batch of companies—including Lumos Robotics, Vbot, Bowintec, Qingtian Zu, Five Ages, and Xynova—secured hundreds of millions in funding at the Series A or Pre-A stages.
Overall, enthusiasm for betting on early-stage projects remains high. But viewed from another angle, this also indicates that the embodied AI sector is still in the early phases of technological exploration and product commercialization.
In terms of sector distribution, the landscape has shifted. Unlike the early days when capital poured primarily into robot manufacturers, funding is now flowing to every core link of the embodied AI supply chain.
Bowintec, for instance, focuses on core components for industrial scenarios like precision assembly and flexible quality inspection. Xynova specializes in dexterous hands, Five Ages develops embodied brains, and Blue Point Control makes six-axis force sensors. None of these are finished-product manufacturers; they are the critical upstream players enabling embodied robots to function.
This shift in capital flows reflects a profound transition underway in the industry: securing an autonomous and controllable supply chain is replacing "building the whole machine" as the new investment thesis.
The logic behind this is straightforward.
For one, following a rush of funding from 2025 through the first quarter of this year, top-tier robot manufacturers are generally well-capitalized.
Moreover, players like Galbot, Spirit AI, Noetix Robotics, GALAXEA, Gigaai, and X Square Robot have each closed at least one 1-billion-yuan round this year. Their valuations have surpassed 10 billion yuan, with some exceeding 20 billion. For new investors, the barrier to entry is now too high—either they can't afford the price tag, or there's no room left in the cap table.
On the other hand, betting on core components offers far more certainty than betting on robot makers. After all, regardless of who ultimately wins the market—be it bipedal or wheeled, industrial or domestic—robots will always need joints, chips, dexterous hands, and sensors.
Image Source: Xynova
This logic mirrors Huawei's decision to become a "supplier of incremental components for smart vehicles" rather than building cars itself. It’s about making money as a "water seller"—betting on the rise of the entire industry, rather than the fate of a single company.
On a deeper level, the shift from funding "whole machines" to "core components" signals that a division of labor within the embodied AI supply chain is beginning to take shape.
When the industry no longer requires every company to build a complete robot from scratch, and instead sees players specializing in joints, sensors, or dexterous hands, it means the sector is moving from a "cottage workshop" phase into an era of "specialized collaboration."
Seen in this light, the current shift in capital flows is merely the shadow cast by this trend on the funding landscape.
Demos Are Losing Value; Investment Logic Is Being Rewritten
The cooling in May appears, on the surface, to be synchronized with a tightening of macro liquidity and a broader correction in the primary market. Therefore, it doesn't signal that the embodied AI sector has entered a downward spiral.
So far, funding activity in the embodied AI space has picked up rapidly again as we move into June.
Observations by Gasgoo show that just in the past few days, multiple companies—including Jianzhi Robot, Heiman Technology, Xingyuan Zhi, Shouyi Technology, Spirit AI, and Stardust Intelligence—have disclosed new funding. Notably, Spirit AI and Stardust Intelligence secured 1.5 billion yuan and 1 billion yuan respectively in their latest rounds.
Viewed this way, the May retreat looks more like a brief pause for breath than a reversal of the trend.
Yet the return of heat doesn't mean the old logic is back in charge. What is truly changing is the investment logic itself.
Since the embodied AI boom began, valuations for early-stage companies were driven largely by "technological narratives." Whose humanoid robot could do a backflip? Whose dexterous hand had more degrees of freedom? Whose large model had more parameters? These metrics directly translated into higher funding rounds and richer valuations.
But now, that logic is failing.
An executive at a leading humanoid robot company summed up current investor screening criteria in three dimensions: commercialization capability, deployment capability, and technological foundation. In plain English: Can you sell it? Can it actually be used? Can it iterate continuously? Without all three, it’s just a "story."
The intensive entry of industrial capital over the past period provides the most direct footnote to this judgment.
Image Source: Bowintec
Industrial giants like Xiaomi, Li Auto, BYD, Geely, CATL, and Bosch are all doubling down on the embodied AI sector. Leveraging their own supply chains, application scenarios, and financial strength, they are deeply deploying resources across both finished robots and core components.
It is worth noting that while industrial capital may be less sensitive to valuations than financial VCs, its requirements for business logic and industrial synergy are extremely high. Crucially, industrial capital brings more than just money—it brings real order scenarios and supply chain resources.
Seen from this angle, companies that secure industrial capital effectively gain multiple advantages at once: investment capital, seed customers, manufacturing partners, and application scenarios.
Conversely, companies lacking industrial deployment capabilities—and without the backing of industrial capital—risk falling behind in the race for funding and mass production, eventually facing market elimination.
This means that the current shift in industrial investment logic is, in essence, a process of survival of the fittest.
Zhang Haixing, founder and CEO of Juzhen Chaozhi, even believes a shakeout in the humanoid robot industry could happen within as little as 18 months. By that logic, by the end of 2027, most of the dozens of humanoid robot startups currently on the market will face elimination.
Chen Tongqing, co-founder of TARS, offers a similar assessment. He argues that the embodied AI industry resembles the consolidation of autonomous driving: initially, a hundred flowers bloom, but eventually, the market concentrates around a few leaders. The rest will either pivot, merge, or dissolve.
The subtext is clear: not every company currently standing in the limelight will be around when the wind actually picks up. When the cheap money stops flooding in, it will become obvious who has been swimming naked.
The Real Battle Begins After 10,000-Unit Mass Production
Although funding cooled briefly in May, the pace of mass production on the industrial side has not slowed. Quite the contrary: several leading companies are densely crossing the same threshold.
Mass production of 10,000 units is becoming the new baseline for top players.
AgiBot Robot is the one running the fastest.
Image Source: AgiBot
At the end of March, AgiBot announced the official rollout of its 10,000th general-purpose embodied robot. The company had only just crossed the 5,000-unit production threshold at the end of 2025. This means it took AgiBot just three months to go from 5,000 to 10,000 units, setting a new record for production ramp-up speeds among domestic general-purpose embodied robots.
Close behind, Zhishen Technology also crossed the 10,000-unit mass production threshold at the end of May, taking roughly five months to scale from 5,000 to 10,000 units.
In early June, Unitree also announced that the cumulative production of a single bipedal humanoid robot model had reached approximately 11,000 units—a figure that does not include other models or wheeled chassis products.
In just three short months, three companies have joined the "10,000-unit Club"—something that would have been hard to imagine just a year ago.
Behind the continuous expansion of production scale by leading companies, the overall production rhythm of the industry is accelerating in sync.
At the end of March, the nation's first automated humanoid robot production line with an annual capacity exceeding 10,000 units—jointly built by Dongfang Precision and Leju Robotics—officially went into operation, rolling out one unit every 30 minutes. In May, Zhongqing Robot's Shenzhen Honghuling base began production, further compressing the rhythm to one unit every 15 minutes.
Embodied AI is moving from the past stage of exploratory small-batch trial production into a true phase of standardized manufacturing.
But a new question arises: as large numbers of robots roll off assembly lines en masse, where will they go?
The answer is not optimistic.
Because real-world application scenarios are numerous and complex, filled with edge cases and non-routine situations, the mass application market for high-performance general-purpose robots is far from mature. Current commercial applications are mainly concentrated in limited scenarios such as quadruped robot inspection, research and education, and consumer entertainment.
Moreover, even in the consumer market, high prices and limited functionality pose severe challenges to large-scale commercialization.
Not to mention that in broader industry scenarios, a complete commercial loop has yet to be formed.
Image Source: Unitree
Unitree's shipment numbers expose this gap thoroughly. In 2025, Unitree shipped over 5,500 humanoid robots—ranking first globally—but 70% of its revenue came from research and education, with industrial applications accounting for only 9%.
Subsequently, Unitree Technical Director Wu Binghao pointed out that for embodied AI to cross its "ChatGPT moment," it still faces three major challenges: first, improving the model's ability to express tasks to break through generalization bottlenecks; second, improving the model's utilization of diverse data to enhance knowledge transfer; and third, increasing the scale effects of reinforcement learning to achieve optimal capabilities across multiple tasks.
These three hurdles cannot be cleared simply by raising money and expanding capacity.
Because the contradiction between 10,000-unit production and 9% industrial orders is essentially not a technological problem, but a mismatch in rhythm. After all, production lines can be built quickly with capital, but trust in industrial scenarios can only be accumulated slowly over time.
This is also the industry's biggest hidden risk right now: capacity utilization matters more than capacity scale.
Therefore, amidst the drumbeat of 10,000-unit mass production, a sharper question is being pushed to the forefront: how far away is the "ChatGPT moment" for embodied AI?
Yang Zhongkai, head of product technology at SFlare Robotics, believes the next five years may be the window for consumer-facing embodied AI deployment, as supply chain maturity and model capabilities are currently converging. In this process, production capacity will be critical: if market demand surges, the ability to quickly match capacity will directly determine who seizes the first-mover advantage.
Chen Wei, VP of Product Ecosystem at Ti5ROBOT, offers an even more optimistic forecast. "I think robots will be everywhere in five years—broadly speaking, scenarios like parks, supermarkets, and roadside shops will see widespread adoption of robot patrol, caregiving, and sales functions, with some companies already deploying these applications at scale. In areas like hazardous zones, industrial parks, community patrols, and power grid maintenance, we will see noticeable changes in the next two to three years. By the five-year mark, robot application scenarios will be much broader."
Furthermore, he believes that in the next five years, owning a robot could become as common as owning a computer today. "Right now, it's relatively difficult for an individual to own their own robot, but I think in the future, the difficulty will be similar to assembling a PC online: you just select the corresponding controller, battery manufacturer, joint manufacturer, sensor components, and shell manufacturer, and you can customize a robot that is exclusively yours."
But beyond the optimistic expectations, we must face a basic fact: the prerequisite for market explosion is that robots must truly be "able to work."
And "being able to work" remains the single hardest hurdle for the entire industry to clear.
Conclusion
Morgan Stanley predicts that by 2036, global adoption of humanoid robots will reach 24.4 million units; by 2050, the global installed base of humanoid robots will hit 1 billion units, with annual market revenue reaching $7.5 trillion.
These figures are enticing enough, but the road to the goal is destined to be a long elimination match. And the funding "cooling" in May can be seen as the prelude to this contest.
When capital stops pouring in frantically, when demos are no longer a core competitive edge, and when 10,000-unit mass production forces the industry into an exam on delivery capability—the survival window for companies stuck at the demo stage, those that can build products but can't open markets, and those that get funding but can't secure repeat orders—is rapidly narrowing.
From "telling stories" to "turning in homework," from "competing for funding" to "competing on delivery," embodied AI is undergoing a rite of passage into adulthood.
