On September 12, the mass production of humanoid robots took a historic step forward as the 10,000-unit-level industrial humanoid robot super smart factory in Liuzhou officially commenced operations. Designed to produce one complete unit every 10 minutes, the facility boasts an annual planned capacity exceeding 10,000 units. This marks a significant transition for China's humanoid robot industry, moving from small-batch trial production into the era of standardized, 10,000-unit-level manufacturing.
Behind this production ramp-up lies a sweeping wave of price reductions across the entire industry. A global physical AI report released by Goldman Sachs in late August provided mid-to-long-term projections: the average price of a complete humanoid robot is expected to decline from $41,800 in 2025 to $21,300 by 2035, a reduction of nearly 49% over the decade. In the Chinese market, however, the pace of price reductions is far outpacing overseas forecasts, as mass production and deployment accelerate to push the price curve downward more aggressively.
Prospectus data from Unitree Robotics reveals a dramatic price journey for its humanoid robots: from 593,400 yuan in 2023 down to 166,400 yuan in 2025, a contraction of nearly 72% in just two years. In June of this year, the consumer-grade product R1 saw its price further reduced to 29,900 yuan, while Songyan Power's compact product "Xiaobumi" has a pre-sale price of less than 10,000 yuan. Market research firm Smart Analytics Global's latest data shows that in the first half of 2026, global humanoid robot shipments reached approximately 19,100 units, a nearly threefold year-on-year increase, with Chinese manufacturers commanding over 97% of the global shipment share.
"Over a ten-year timeframe, a 50% decline in average unit price is not aggressive; for certain technology routes, it may even be conservative," said Zhang Zhengtao, founder and chairman of core component supplier Zhongke Huiling. However, a prevalent misconception persists in the market: equating the price decline of embodied products directly with the arrival of the industry's commercialization inflection point. After conversations with multiple practitioners in embodied intelligence working on the front lines, it becomes clear that the halving of product prices fundamentally signals the industry merely crossing the entry threshold. The core elements that truly trigger large-scale industry explosion have never been about low prices alone.
Three Forces Converge for Industrial-Level Cost Reduction
This round of price cuts is not a short-term promotional tactic, but rather an industrial-level cost decline driven by the convergence of scale effects, domestic substitution, and technological iteration. First, large-scale production dilutes fixed costs. According to Zhang Shaozheng, co-founder of Yuanli Lingji, current production volumes at leading manufacturers have approached or reached the 10,000-unit level, creating preliminary scale effects. Under this scale, while costs have not yet reached their floor, they are at an acceptable range, and component consistency has improved. Beyond the Liuzhou facility, Unitree's IPO fundraising projects also include an annual production capacity of 75,000 humanoid robots. As procurement volumes rise, component pricing power, mold amortization, assembly efficiency, and yield rates continue to improve, replicating the classic manufacturing logic of scale-based cost reduction.
Second, the localization of core components is breaking down the price system. According to data from the China Academy of Information and Communications Technology, the overall localization rate of core humanoid robot components has surpassed 75%, with Morgan Stanley's calculations suggesting a localization rate of over 90% for certain core components. Without China's supply chain, the cost of a single unit would skyrocket to $131,000, a nearly threefold difference. The localization process is not uniform across all segments: motors and controllers have localization rates exceeding 85%, while high-end components like six-dimensional force sensors, which were once less than 30% locally sourced, have rapidly achieved domestic substitution breakthroughs in recent years. Even so, core hardware prices have experienced a landslide decline: harmonic reducers have dropped from 3,000-5,000 yuan to approximately 1,500-2,000 yuan, planetary roller screws have seen reductions of over 80%, and six-dimensional force sensors are moving from the 10,000-yuan level toward below 5,000 yuan, solidifying the supply chain foundation for overall unit price reductions.
Third, the convergence of technology routes is coupled with manufacturers' strategic market positioning. An industry insider in embodied intelligence noted that the technology routes are becoming increasingly clear, with redundant hardware designs being phased out in favor of highly integrated joints and lightweight structures. Leading manufacturers are using low-priced models as ecosystem entry points, covering scientific research and educational scenarios, accumulating real physical data to refine algorithms, and trading short-term profits for long-term ecosystem moats.
Zhang Shaozheng's assessment goes further: the speed and ceiling of domestic unit price reductions will likely outperform overseas markets, but the scale-up milestone may not necessarily arrive earlier—because price reduction and mass production are two separate issues. Price reduction relies on the supply chain; mass production relies on the brain. Supply chain advantages address whether products can be made cheaper, while production scale determines how thin costs can be spread. What truly determines whether the market opens its wallet is the maturity of the embodied brain. In his view, the former represents an existing advantage, while the latter is the decisive variable.
The Hardest Cost to Cut
Breaking down the complete unit's BOM structure, actuator systems account for approximately 45% of costs, and sensor systems around 15%, together exceeding 60% of the total. The areas of most concentrated cost also happen to be where domestic localization is still climbing. Cost reduction potential remains, but structural bottlenecks are equally prominent. One of the most critical shortcomings is the dexterous hand. Goldman Sachs lists dexterous hands, system reliability, and real-world scenario data as the industry's three major constraints. Currently, high-degree-of-freedom tactile dexterous hands range from over 100,000 to several hundred thousand yuan per unit, yet their service life is only a few weeks to two or three months. The cheaper the complete unit becomes, the higher the cost proportion of the dexterous hand, making it a core bottleneck for deployment.
As a dexterous hand supplier, Zhang Zhengtao does not believe this ceiling will persist indefinitely. "The high price of many high-end dexterous hands today is essentially the combination of three factors: high degrees of freedom, high integration, and low production volumes," he said. Over the next three years, dexterous hands will remain in a clear cost-decline cycle, driven not by "materials suddenly becoming cheaper," but by design cost reduction, process optimization, supply chain improvements, and scale effects. Zhongke Huiling's basic dexterous hand has already achieved a price point of several thousand yuan. However, he believes the ultimate cost limit is determined by the minimum performance and reliability required to accomplish target tasks. "If you cut a few hundred yuan but significantly reduce lifespan, precision, or task success rates, that's not cost reduction—it's shifting costs to maintenance and downtime."
Even more easily overlooked are the hidden costs. Zhang Shaozheng points out that industry discussions of cost are almost entirely confined to the hardware BOM, while long-term investments in software R&D, model iteration, and data accumulation are not factored in. Hardware cost reduction has a clear physical limit, with the curve flattening after five years. But the efficiency cost reductions brought by AI intelligence have no ceiling—the stronger the model, the more scenarios a single unit can cover, the lower the proportion of manual fallback needed, and unit operating costs can be continuously spread thinner without hardware modifications. In short, hardware solves affordability; intelligence solves usability. The former is something everyone will eventually achieve; the latter is only attainable by companies that truly possess models, scenarios, and data.
Will Price Cuts Compromise Quality?
As prices rapidly decline, signs of a price war have emerged, bringing industry bubble concerns and quality risks to the forefront. Since the first quarter of this year, the price war has spread from consumer-grade to industrial-grade products: Unitree's G1 has dropped to 85,000 yuan, Accelerated Evolution's K1 is promoted at a limited-time price of 29,900 yuan, Xingchen Intelligent's industrial-grade product T1 has been priced at 89,900 yuan, marking the first time industrial humanoid robots have entered the sub-100,000 yuan range. In late June, UBTech released its full-size humanoid robot U1 series, starting at 119,800 yuan with pre-sale orders exceeding 13,000 units.
Yet there is a mismatch between capital enthusiasm and industrial deployment. In the first quarter of 2026, domestic financing in this sector exceeded 68.1 billion yuan, but global shipments in 2025 were only approximately 17,000 units. Among the more than 170 related enterprises, most remain at the stage of "having demonstrations but no mass production," with some early engineering prototypes even being liquidated at low prices on second-hand platforms. More concerning is the divergence in the nature of price reductions. Zhang Shaozheng categorizes them into two types: price reductions driven by scale effects, yield rate improvements, and process optimization naturally do not compromise performance—in fact, the more units shipped, the more failure modes are exposed and addressed, making reliability more robust. It is only the cost-cutting and material downgrades forced by price wars that degrade quality. "Quality degradation from price cuts is not an inevitable outcome of lowering prices; it's an inevitable outcome of lacking real-world scenario data—companies without large-scale deployment don't know which cost cuts will cause problems, so they can only cut blindly." He draws a clear red line: reliability is not negotiable. Robots are not consumer electronics; they are means of production. A robot that breaks down every few days is meaningless even at a 20,000 yuan discount. Performance can be tiered, but reliability has no high, mid, or low end.
Zhang Zhengtao also cautions that the robotics industry is still in a phase of rapid technological and product evolution. In this phase, the most important thing should be improving task completion rates, reliability, and product value. If technology capabilities are not yet mature, entering pure price competition prematurely can shift the entire industry's focus from "solving problems" to "reporting a lower number." "We believe the robotics industry should ultimately compete on unit task cost, not merely on equipment selling price."
The Inflection Point Isn't Price
So what constitutes the true industry inflection point? The answers from multiple interviewees are strikingly consistent: the critical threshold of usefulness. "Currently, production capacity, costs, and the supply chain have basically established the commercial foundation; the core constraint remains practical capabilities," Zhang Shaozheng said. When robots cross this threshold, orders will follow naturally without anyone needing to make predictions; if they cannot cross it, production capacity will only become inventory. He notes that customers calculate the total ROI equation: a robot with a 99% success rate versus one with 95%—the four percentage point difference looks small, but the actual manual fallback costs differ by several times. "The cheap robot ends up being the most expensive."
As for the pricing system, he predicts that the next decade will not follow the smartphone model of "one size fits all," but rather resemble the automotive industry's tiered competition where "from Wuling to Rolls-Royce coexist," with each price band developing its own market leaders. Zhang Zhengtao provides an industrial timeline: 2027 to 2028 will be the critical validation period, focusing on whether robots can stably implement standardized processes in industrial, logistics, and other scenarios, completing the leap from "demonstrable" to "functional." If supply chains, model capabilities, and production progress maintain their current pace, 2028 to 2030 will become the true cost and scale inflection point, as the industry transitions from project-based small-batch procurement to standardized commercial procurement. Consumer mass markets will arrive later, as home scenarios demand far higher safety and reliability standards than industrial workstations.
Returning to the industrial node of the 10,000-unit factory coming online, capacity release will further accelerate price declines. Goldman Sachs' decade-long "halving" prediction will likely materialize, or perhaps even arrive ahead of schedule. But as Zhang Zhengtao articulates: what ultimately determines the speed of adoption is never the difference between $20,000 and $30,000. The true inflection point arrives when robots can consistently complete high-value work every single day—at that moment, their price truly becomes affordable.