The "Schism" in Humanoid Robotics: Hype Goes Viral, But Practical Application Has "Short Legs"

Gasgoo Munich- A subtle sense of "schism" hangs over the humanoid robot sector in 2026.

On one side, the hype keeps building: smooth demos at trade shows are going viral, companies are piling into the space, and forecasters keep revising their shipment projections upward, with some now pegging full-year global deliveries at nearly 100,000 units.

On the other side lies the public's direct experience: forget about entering ordinary homes—even on factory floors, it's rare to see humanoid robots actually filling shifts or replacing human workers.

Just how wide is the gap between these flashy numbers and the quiet reality on the ground?

"I don't expect to see clearer progress on true mass production and scaled pilot applications until at least the second half of next year," says Wang Xianbin, a partner and vice president of research at Gasgoo.

Deliveries are set to surge, but large-scale deployment is still waiting in the wings. That time lag is the key to understanding the industry's current state.

How much of those 100,000 shipments is real end-market demand?

If you look only at the shipment figures on paper, humanoid robots seem to be on the verge of a mass-market explosion.

Latest data from U.S. market research firm Smart Analytics Global (SAG) shows global humanoid robot shipments reached roughly 19,100 units in the first half of 2026—more than triple the 5,100 units from a year earlier. SAG forecasts full-year global shipments will hit around 60,000 units, climbing to 500,000 by 2030.

Gasgoo Automotive Research Institute offers an even more optimistic take: this year's global shipments could approach 100,000 units.

But Wang quickly points out that current shipments don't necessarily equal real end-market demand. A significant portion comes from three types of "non-terminal demand."

Image Source: AgiBot

The first category is related-party transactions and ecosystem mutual procurement.

With humanoid robots still in a phase of technological iteration, early delivery orders from many leading companies come primarily from internal purchases by shareholders or supply chain partners.

The core purpose of these purchases is rarely to generate revenue through actual production. Instead, they're used for internal pilot testing, technical verification, or brand image building.

For robot manufacturers, this is a necessary path to stabilize cash flow and validate product iterations in the early stages. But essentially, it's an internal loop within the ecosystem—not a reflection of genuine willingness to pay in the end market.

The second category is research and secondary development procurement.

As embodied intelligence heats up, large model developers, university labs, and research institutions have become major buyers.

Their core motivation for buying the hardware is to use it as a carrier for algorithm training, data collection, or scenario validation. Essentially, this is R&D investment, not productivity procurement, and it lacks scenario replicability.

The third category is channel stocking and display rentals.

Most humanoid robots the general public encounters are found in science museums, shopping complexes, brand launch events, or trade shows. They handle greeting, performing, or interactive guiding—offering "display value" rather than strict production or service value.

Because of this, the demand structure for humanoid robots shows a clear "inversion."

Sales data from Gasgoo Embodied Intelligence reveals that commercial service applications account for nearly 90% of current shipments. Entertainment and performance displays alone make up 35%, while data collection and research account for 20%. The rest is split among university research, business reception, and other commercial services. Orders actually used in production to create direct industrial value make up less than 15%—far from forming an indispensable, rigid value proposition.

Yet industrial production is precisely one of the markets with the most urgent need for humanoid robots.

"Our user surveys show that industrial scenarios are where expectations are highest—tasks like process operations, quality inspection, equipment maintenance, and safety patrols," says Li Junlan, a research manager at IDC China. "Conversely, users see less urgency in areas like health maintenance or entertainment performance."

High expectations for industrial replacement on one hand, and a sub-15% deployment rate in production on the other—this gap is pushing the industry to accelerate validation in real-world scenarios.

In automotive manufacturing, for instance, the wire harness threading operation developed by TARS in partnership with Tianhai has achieved continuous process validation on the production line. Xiaomi's humanoid robot is already working in its own EV factory, handling tasks like loading self-tapping nuts and moving totes. In 3C manufacturing, AgiBot's Elf G2 robot has been deployed on Longcheer Tech's production line, where it has been "interning" for over eight months. In warehousing and logistics, StarDance Epoch's StarDance M7 is now in routine operation at logistics centers for SF Express and China Post.

Even so, these pilots remain in the "feasibility validation" stage. The industry has yet to see large-scale market repurchase orders, and no scenario has truly produced a clear, replicable ROI model.

There is still a long way to go from "can do" to "worth doing."

Why is it so hard for humanoid robots to "get a job on the factory floor"?

The inverted demand structure is just the symptom. The root cause is that humanoid robots haven't crossed three core barriers to scaled deployment: the data dilemma, process adaptation, and hardware costs.

Each of these challenges is a tough nut to crack, and they are deeply interconnected.

First and foremost is the data dilemma.

"For embodied intelligence, the gap between high user expectations and actual product capability comes down to a lack of autonomy," Li Junlan points out. "Robots still can't autonomously assess the state of the physical world or seamlessly combine commands with the real environment. So the biggest challenge right now lies in algorithms and models—and the core 'fuel' for iterating those models is data."

Yet deployment scenarios for humanoid robots are currently extremely limited, leading to a severe shortage of scenario-specific data for training. Although many companies have started data collection, the process itself is labor-intensive. The industry hasn't solved the problem of data standardization, meaning the collected data is often fragmented and unstructured—making it hard to form high-quality, reusable training sets.

This is especially true in real industrial settings, where acquiring effective data is incredibly expensive. You have to ensure the production line runs without interference while covering enough variables and edge cases. That places stringent demands on the systematic nature, safety, and representativeness of data collection.

The lack of unified data interfaces, labeling standards, and sharing mechanisms exacerbates the "data silo" problem. Companies are forced to duplicate efforts and fight their own battles, unable to accelerate model evolution through scaled collaboration.

This negative cycle—where a lack of data hinders capability improvements, which in turn blocks real-world validation, which then prevents access to high-quality data—is becoming the key shackle holding humanoid robots back from "demo-ready" to "stable and reliable."

Image Source: TARS

Second is the mass production barrier: demos prove whether you *can* do it, but mass production asks whether you can do it *consistently well*.

For a humanoid robot to truly "work" in a factory, the demands on hardware precision, motion latency, sensor perception, and system stability are incredibly high. Moreover, manufacturers must deeply understand the specific industry's workflow processes, not just hand over the hardware.

"Take visual inspection, dispensing, or assembly on the line, for example. These often require the robot to achieve precision within a few hundredths of a millimeter at a fixed station," Wang Xianbin explains. "The pace on the final assembly line is very fast. Once a part arrives, the robot can't wait a few seconds to start. Industrial scenarios demand extremely high real-time response capabilities."

The reality, however, is that robot manufacturers excel at hardware design but lack deep understanding of vertical industry workflows. Meanwhile, industry clients who master the processes aren't familiar with the technical boundaries of robots. This information gap means that deploying a robot in any new scenario requires a long period of adaptation and debugging, driving up time and labor costs.

For a factory, the criteria for judging a robot's value are simple: efficiency and cost.

If a robot replaces a human, how long can it run continuously without failure? What is the payback period? How much maintenance is required over its lifecycle? Do engineers need to be stationed on-site permanently if it breaks down? If these questions can't be answered, entering the factory is just empty talk.

Finally, there is the hardware cost knot: without economies of scale, prices won't come down.

Since the start of the year, calls for lower prices have been constant. Noetix Robotics' "Xiao Bumi" even dropped below 10,000 yuan, pulling the entry barrier down to a level mass markets can touch. But there's no denying that such products are largely for early adopters and are meant to cultivate market awareness.

For true production-grade humanoid robots, however, prices generally hover near 100,000 yuan or higher. Unitree's G1, ENGINEAI's PM01, and Astribot's T1 all start between 85,000 and 90,000 yuan. Chery's AiMOGA M1 and Leju's Kuavo are priced higher, at 285,700 yuan and 308,100 yuan, respectively. High-end models easily exceed 400,000 yuan.

Behind this lies the rigid constraint of hardware cost structures.

"Looking at cost structure, the actuation system accounts for the largest share," says Wang Xianbin. "Core components like harmonic reducers, coreless motors in dexterous hands, and encoders make up about half of the total cost. Perception parts like six-axis force sensors and LiDAR account for roughly 15% to 20%. The rest includes batteries, chips, and other structural components."

This means the hardware cost of the actuation system alone locks in the price floor for the entire machine.

There is a bright side, however. Wang believes the cost curve for humanoid robots will closely resemble that of the auto industry. As shipment volumes expand and the supply chain matures, hardware costs will naturally fall. Most core hardware could see price drops of 40% to 70%. Components like six-axis force sensors, dexterous hands, and LiDAR have significant room for cost reduction—some could see their costs cut in half.

"By 2030, the cost of humanoid robots for scaled applications is expected to drop to $30,000 to $40,000," Wang says. "At that point, if product capabilities mature in parallel, they'll have the chance to enter more commercial scenarios and even gradually move into the home market."

Beyond the three mountains of data, process, and cost, there is one easily overlooked barrier to large-scale deployment: safety.

Image Source: ENGINEAI

Given current trends, deployed humanoid robots will inevitably coexist with humans, entering our work and living spaces directly. If a robot were to suddenly lose control and kick or hit someone, the consequences could be severe. There have already been incidents where humanoid robots injured R&D personnel during testing.

Information security cannot be ignored either. Robots will converse with users and collect environmental data via cameras and microphones. Ensuring this data isn't leaked, preventing robots from outputting malicious content, and stopping them from generating speech that harms public or national security are all issues that must be solved.

As the auto supply chain crosses over, how can it differentiate its strategy?

Despite the deployment challenges, the application scenarios are broad enough that beyond embodied intelligence startups, many companies are expanding into the space. Among them, companies from the automotive supply chain are a force that cannot be ignored.

A review by Gasgoo Embodied Intelligence shows that among automakers, top players like Xiaomi, XPENG, Li Auto, Chery, GAC, Dongfeng, and FAW are actively pushing humanoid robot R&D. On the parts side, companies including Joyson Electronics, Desay SV, Tuopu Group, Sanhua Intelligent Control, RoboSense, and Black Sesame Technologies are all scrambling to stake their claims.

In Wang's view, there are three main reasons why auto suppliers are flocking to humanoid robots. First, many automotive components and manufacturing capabilities can be repurposed. Second, the auto industry has already built mature engineering, supply chain, and mass production systems that can spill over into robotics. Third, as competition in the auto sector intensifies, companies are looking for new growth curves.

But similar motivations don't mean identical paths. As everyone rushes in, how can they build a differentiated advantage? Wang suggests four strategic paths.

The first path is from auto components to robot hardware: direct application of technical accumulation.

The auto industry's expertise in sensors, actuators, thermal management, and electrical/electronic architectures is highly compatible with humanoid robots. Automotive-grade reliability standards also naturally align with the stringent requirements of industrial robots. This means many parts suppliers can fast-track their entry into the core robotics track without starting from scratch, securing key positions in the supply chain.

Joyson Electronics is a prime example. Its focus on embodied intelligence "brains," electronic skin, and solid-liquid hybrid batteries is essentially a cross-border extension of automotive electronics and powertrain technologies, quickly building technical barriers around core components.

The second path is dual output of scenario and manufacturing: bridging the industry's process gap.

Another unique advantage of the auto industry is that it brings its own real-world deployment scenarios and a mature manufacturing system.

On the scenario side, automakers' own factories, offline stores, retail networks, and after-sales systems are the perfect natural testing grounds. Automotive production lines, in particular, feature high standardization, large labor volumes, and high repeatability—making them ideal for small-batch validation and rapid iteration of humanoid robots.

Image Source: Xiaomi Technology

On the manufacturing side, the auto industry's mature supply chain management, quality control standards, and mass production capabilities can provide contract manufacturing and engineering support to robot makers. This solves the pain points of robot manufacturers who lack mass production experience and weak supply chain management skills.

Automakers can therefore act as "scenario definers," opening up their process expertise to co-define products and validate solutions with robot manufacturers. This bridges the current core gap where "robot hardware is disconnected from industrial processes."

The third path is entering the embodied intelligence and AI terminal markets directly.

Some automakers and suppliers are already cutting directly into these tracks, developing robots, smartphones, AI glasses, and other intelligent terminals. This approach essentially extends the automotive industry's existing platform technologies, R&D systems, capital, and talent resources into new product areas.

Comparatively, this path requires heavier investment. But for leading companies with the capital, technology, and resources, it could become a crucial way to find a "second growth curve."

The fourth path is cross-border collaborative innovation centered on the automotive ecosystem.

A car is not just a vehicle; it can also be seen as a mobile energy node, computing node, and space node. In the future, robots could collaborate more deeply with cars to create new scenarios that never existed before.

For example, from a technical perspective, quadruped robots could serve as mobile power sources or even mobile edge computing nodes. Like an AI Box, they could supplement a vehicle's perception range and computing power, collect environmental data outside the car, and support large model training and scenario applications.

In this sense, the entry of the automotive supply chain isn't just simple cross-border "bandwagon-jumping." It's bringing mature manufacturing systems, scenario resources, and supply chain capabilities to the table—injecting real momentum into an industry stuck at the mass production threshold.

These moves may not immediately rewrite the industry landscape, but they are likely to become the key external variables driving humanoid robots from validation-level shipments to scaled deployment over the next two to three years.

In Conclusion

At the end of the day, the "froth" in today's humanoid robot market isn't worthless bubble—it's a necessary stage in the early life of any emerging industry.

From an industry perspective, any disruptive hardware goes through the full cycle of "concept hype—pilot validation—cost decline—scaled deployment." Humanoid robots are no exception.

Ultimately, the companies that truly survive the cycle may not be the ones with the flashiest demos. They will be the ones that dive deep into industry scenarios, stabilize their products, drive down costs, and prove the ROI.

After all, for industrial products, the most hardcore sense of technology never comes from a one-minute performance on stage—it comes from day-after-day stability on the production line.