Gasgoo Munich-Although the humanoid robot industry has only been exploding for two years, the market is already facing violent price swings.
Cast your mind back to 2024. A full-size humanoid robot cost hundreds of thousands or even millions of yuan—a luxury out of reach for most companies and individuals. The industry was still in the lab verification phase, with global shipments under 2,000 units and a supply chain far from maturity.
Today, just two years on, the market landscape looks completely different.
In the high-end market, "all-rounder" models like Unitree H1 and Matrix Super Intelligence MATRIX-3 still hold the line above 500,000 yuan, representing the industry's cutting edge. At the opposite extreme, Noetix Robotics' "Xiaobumi" has dragged the entry barrier down to within reach of the mass market with a price tag under 10,000 yuan.
From hundreds of thousands to under 10,000 yuan, humanoid robots have completed the downward price trajectory from "lab luxury" to the prototype of a "consumer product" in just two years.
Is this apparent "avalanche" of price cuts the inevitable result of a maturing supply chain, or a sign of a bubble inflated by capital?
Price "Avalanche"
Unitree offers the clearest snapshot of this current price dive.
According to Unitree's prospectus, the average unit price of its humanoid robots was nearly 600,000 yuan in 2023. By 2024, it had plunged to 260,000 yuan—a 57% drop—and in 2025, it fell further to 166,400 yuan.
Image source: CCTV Spring Festival Gala screenshot
Behind these figures: Unitree's first full-size humanoid robot, the H1, launched in 2023, cost 650,000 yuan. But just a year later, its second humanoid bot, the G1, was priced as low as 99,000 yuan. In 2025, Unitree dropped another rung; the R1 series launched that year started at just 29,900 yuan for the AIR version.
It is worth noting that while the G1 and R1 saw massive price cuts, they also represent generational performance gaps compared to the H1. The H1 was built for industrial frontiers and extreme sports, equipped with a high-explosive power system, maximum joint torque of 360 N·m, peak torque density of 189 N.m/Kg, an 864 Wh battery, and a "3D lidar + depth camera" fusion perception system.
The G1 serves as the middle ground: smaller at about 1.3 meters tall and weighing roughly 35 kilograms. Its athletic ability is relatively weaker, placing more emphasis on fine manipulation skills.
The R1 focuses on extreme lightness. It is even smaller than the G1, and because it prioritizes the arms themselves, it lags behind the G1 in full-body coordination and fine manipulation.
Image source: Unitree
This shows Unitree's price cuts weren't sudden slashes on the same product, but rather a strategy of using different performance tiers to meet broader market demand.
Nevertheless, the downward trend in humanoid robot pricing is undeniable.
Leju provides a direct example. According to its prospectus, the company's full-size "Kuafu" series sold for about 414,000 yuan per unit in 2024, but dropped to 308,000 yuan in 2025—a 25.6% annual decline.
Across the wider market, humanoid robots are shattering previous price floor expectations.
Recently, Stardust Intelligence launched the Astribot T1, a new series of tendon-driven AI robots. While maintaining the precision and speed advantages of tendon transmission, the starting price is only 89,900 yuan. Previously, the company's research-grade S1 robot sold for about 500,000 yuan.
Accelerated Evolution's "Embodied Development Entry-Level Platform Booster K1," released in October 2025, starts at just 39,900 yuan. At launch, the limited-time price even dipped to 29,900 yuan.
Noetix Robotics' "Kid Brother" Bumi (Xiaobumi) pushed the price below 10,000 yuan. With a limited-time price of 9,998 yuan, it became the industry's first mass-produced consumer humanoid robot under the 10,000 yuan mark.
Image source: JD.com screenshot
However, interpreting "price cuts" as simply meaning all humanoid robots are getting cheaper is too one-sided. A more accurate observation is that the market is undergoing a profound price stratification, shifting from a "one-size-fits-all" research pricing model to a diverse landscape with clear positioning and distinct tiers.
The first tier: "Entry-level tasters" under 40,000 yuan, represented by Noetix Robotics' Xiaobumi, Unitree R1, Accelerated Evolution's Booster K1, and Zhenqiong SA01.
These models are typically smaller with limited degrees of freedom. They lower costs to cover education, entry-level research, and light home companionship, while also cultivating market awareness.
Take Xiaobumi: its 94-centimeter height and 12-kilogram weight mean it cannot handle complex industrial tasks, but it suffices as a feature-rich home companion and programming education platform.
The second tier: "Productivity tools" between 40,000 and 300,000 yuan. This is the most fiercely competitive segment with the richest selection, including Unitree G1, Stardust Intelligence T1, Zhenqiong PM01, Zhenqiong T800, Chery Mojia M1, and Leju Kuavo.
Unlike entry-level models, robots in this range generally feature full-size or near-full-size designs, with stronger load capacity, longer battery life, and higher degrees of freedom. They are no longer just "toys" or "teaching aids" but are starting to deliver real value in commercial services and industrial manufacturing. This is the core battleground where companies are fighting for customers and pushing technology into real-world scenarios.
Image source: Matrix Super Intelligence
The third tier: All-around "technology flagships" over 500,000 yuan, represented by Unitree H1 and Matrix Super Intelligence's MATRIX-3. The standard MATRIX-3 costs 580,000 yuan, while the PRO version is 680,000 yuan.
These products often represent the manufacturer's peak technical capability. By maxing out specifications, they achieve superior performance and serve as vehicles for technical exploration and frontier validation.
Take the MATRIX-3: it features an NVIDIA AGX computing platform, 3D biomimetic fabric skin, the MATRIX HAND dexterous hand, and automotive-grade batteries offering up to 4 hours of range. For context, most humanoid robots currently average between 2 and 4 hours.
However, this also implies these products still have a long way to go before true large-scale commercialization.
The Truth Behind "Dirt-Cheap" Prices
This round of price cuts for humanoid robots is not a simple internal "price war," but the result of a maturing domestic supply chain combined with the release of scale effects.
Domestic substitution of core components is the first "cornerstone" supporting cost reduction.
Leju Robots notes that the 2025 average selling price of its Kuafu series dropped 25.56% year-on-year, mainly due to continuous cost optimization: pushing domestic production of core parts, optimizing structural topology, and supply chain collaboration. Meanwhile, the maturing technology and performance of upstream suppliers, along with emerging scale effects, created favorable conditions for lower costs.
In the past, core components like reducers, servo systems, force sensors, and encoders were heavily reliant on imports, driving up costs. Now, the wave of domestic substitution is rewriting the landscape.
Take harmonic reducers: analysis shows imported brands like Japan's Harmonic Drive typically cost between 2,000 and 3,000 yuan per unit, while domestic products range from 800 to 1,500 yuan—a significant price advantage.
Furthermore, analysts predict that as scaling progresses, there is still substantial room for harmonic reducer prices to fall.
Image source: Lingxin Qiaoshou
Dexterous hands are another prime example. Lingxin Intelligent's high-performance DexHand021 Pro costs roughly 14,000 to 28,000 US dollars while maintaining top-tier specs—just one-fifth the price of international competitors.
Through self-developed core components and supply chain optimization, Lingxin Qiaoshou has undercut foreign rivals significantly while maintaining comparable performance—sometimes to a fraction or even a tenth of the price. For instance, the company's Linker Hand O6 is priced at just 6,666 yuan, less than one-twentieth of similar international products.
The dawn of mass production for humanoid robots provides another engine for sustained cost decline.
Around 2024, most humanoid robot companies shipped only dozens to hundreds of units annually. The industry operated more like "workshops," with each robot requiring extensive custom production and debugging. This is why many robot manufacturers opted for full-stack in-house core components, naturally driving up costs.
By 2025, the situation had fundamentally changed.
Data from several analysis firms indicates global humanoid robot shipments in 2025 landed between 13,000 and 18,000 units, with nearly 90% coming from Chinese companies.
Entering 2026, leading enterprises are sprinting toward higher production volumes. Zhiyuan, Zhishen Technology, Unitree, and others have already crossed the 10,000-unit mass production threshold. Meanwhile, Leju, Zhenqiong, and others are actively expanding capacity and building standardized production lines.
Exponential growth in shipments—from hundreds to tens of thousands—has dramatically lowered the R&D, mold, and line depreciation costs allocated per unit. Scale effects are beginning to show.
Technological cost reduction cannot be ignored either.
Noetix Robotics' Xiaobumi, for instance, broke the 10,000 yuan barrier by using extensive composite materials to balance light weight with low cost. Additionally, deep in-house R&D on controllers and motor systems helped drive down costs.
Image source: ENGINEAI
ENGINEAI also leverages full-stack self-development as a key cost-cutting tool. Early plans aimed to control humanoid robot prices between 150,000 and 200,000 yuan through full-stack control. Currently, its latest full-size T800 has entered that range, with the base version priced at 180,000 yuan.
Chen Jianyu, founder of Robot Era, also noted that the industry currently lacks high-performance standardized hardware. Full-stack self-development not only allows for rapid tech iteration and flexible model adaptation but also effectively controls hardware costs, laying the foundation for scaling.
Some companies are even exploring software algorithms to compensate for hardware precision limitations, reducing reliance on expensive components.
Of course, beyond these factors, some companies are actively sacrificing margins to seize first-mover advantage.
Unitree explicitly stated that with increasing entrants and intensifying competition, it adjusted prices downward in 2025 to consolidate its position and market share. The decision came after assessing market development, cost trends, and profit margins.
After all, the industry is in a critical "land grab" phase. Whoever first accumulates users, acquires scenario data, and builds a developer ecosystem is more likely to dominate the industry's explosive growth phase.
The Eve of a Shakeout
What does this massive price drop mean for the humanoid robot industry? The answer isn't a simple "good" or "bad," but a complex picture of intertwined opportunity and challenge.
The most immediate benefit of this price war is undoubtedly the lower barrier to entry, accelerating the formation of an industry closed loop.
On one hand, lower prices allow more SMEs, startups, and developers to purchase robots for secondary development and scenario adaptation at a lower cost. This enriches the application ecosystem and drives faster tech iteration.
On the other hand, affordable prices give ordinary consumers a chance to experience humanoid robots, cultivating C-end market awareness and laying the user foundation for future mass commercialization.
Notably, this round of price cuts has spilled over into the rental and second-hand markets.
Image source: Qingtianzu screenshot
Daily rental rates have dropped from tens of thousands of yuan in early 2025 to just a few thousand. Second-hand research prototypes have even seen clearance prices of "50,000 yuan per carload." Lower experience barriers give more potential users access to the technology, pushing the industry from R&D toward scaled commercial use.
However, a warning is in order: when the price bubble is squeezed out, the gold content of value becomes the only "touchstone."
For the past two years, the humanoid robot track focused on "can it stand up," "can it do a backflip," and "how many degrees of freedom." These metrics demonstrate technical strength but remain a layer away from the actual needs of business users.
When a robot costs 650,000 yuan, a client might pay a premium for "technological exploration." But when entry-level products drop below 30,000 yuan, the rational buyer will inevitably ask: How much money can this machine actually save me?
This isn't nitpicking; it is the inevitability of a maturing market.
Currently, humanoid robots are transitioning from "lab favorites" to "productivity tools." This shifts the competitive focus from flashy athletic moves to ROI and even repurchase rates.
Companies that cannot answer this question will be eliminated, no matter how cool their technology is.
From this perspective, the price war signals the start of an industry shakeout. The humanoid robot sector remains incredibly capital-intensive. Aside from a few like Unitree, most are deep in the red. Cutting prices further compresses margins, worsening losses.
Moreover, considering the industrialization process is still in its infancy, honing core technology is more critical than price cuts.
"Given the current state, embodied intelligence in 2026 will only manage small-scale commercial pilots in single, simple scenarios. Mass deployment requires long-term tech iteration and scenario polishing," said Zhu Xing, CEO of Ant Lingbo Technology, at the recent Beijing Academy of Artificial Intelligence conference.
Zhu added that physical AI still faces hurdles in physical perception, dynamic interaction, and generalization. In physical perception, robots have vast room for improvement.
"In terms of vision, enabling robots to see clearly and perceive accurately is a long-term core challenge. Depth perception and distance judgment remain weak points. In touch, we need high-precision sensing, but more importantly, we must integrate tactile data into model inference, breaking the limitation where touch is only used for shallow end-effector control. Beyond that, multimodal perception like sound and temperature presents many technical problems to solve," Zhu stated. Additionally, under current model architectures, insufficient generalization remains a core challenge.
Even though hardware and "cerebellum" control have iterated rapidly in recent years, Zhu believes this doesn't mean they are fully mature. "The AI brain's intelligence upgrade will redefine the hardware system, starting from perception and driving a comprehensive new transformation in hardware."
Dong Kai, director of the Technology Department at the CCID Institute under the Ministry of Industry and Information Technology, was even more blunt: starting a "price war" at this stage isn't a phenomenon to be encouraged.
"This year, the BOM cost of embodied robots is clearly dropping. For joint modules, BOM costs have fallen 30% to 40%. Foreign institutions predict these costs could drop 80% to 100% in two or three years. But our output hasn't risen significantly yet, so we haven't reached the stage where we can support rapid marginal cost reductions," Dong said.
Therefore, behind the talk of "dirt-cheap" robots, we need to see clearly that price cuts are a phased phenomenon. True mass adoption still requires breakthroughs in core technology, scenario adaptation, and production capacity support.
Price cuts may have opened a lower threshold for the industry, but how far it ultimately goes depends on the joint accumulation and deep cultivation of the entire supply chain.
