Is China's robotics bubble about to burst? The "valley of death" of robotics: China's radical plan for humanoid robots

Not the factory, but the data: This is how China's robotics startups really make their money

Bubble warning: What the West must now learn from China's robot strategy

Billions in investment, a gold rush mentality, and suddenly an official warning about a bubble: China's humanoid robot industry is at a crucial turning point. While startups are raising capital at a breathtaking pace and presenting new prototypes almost weekly, practical testing in factories reveals a harsh reality. The technology is simply not yet ready for fully autonomous mass production. The industry is currently traversing the dreaded "valley of death"—that critical phase in which research funding dries up, but real market revenues still seem a long way off. To survive, the smartest Chinese manufacturers are undergoing a radical change of strategy. Instead of waiting for the perfection of factory AI, they are generating urgently needed cash flow through entirely new business models: They are renting out their robots for events, building gigantic centers for collecting valuable training data, and occupying highly lucrative but less complex niches in high-risk professions. This pragmatic survival strategy not only ensures the survival of individual companies, but could also give China a structural data and market advantage that is almost impossible for the West to overcome.

Technology alone is not enough – those who don't earn money die

Between gold rush mentality and regulatory disillusionment

China's humanoid robotics industry is in a state of productive schizophrenia. On the one hand, capital is flowing in at a pace unusual even by Chinese standards: In 2024 alone, the sector raised an estimated 48 billion yuan (around US$6.7 billion) across 89 funding rounds. On the other hand, in November 2025, the National Development and Reform Commission (NDRC), China's top economic planning agency, publicly warned of a bubble – a rare move for an agency that typically acts as a promoter of strategic industries. Spokeswoman Li Chao succinctly summarized the problem: Over 150 companies are producing humanoids, and more than half of them are either newly founded startups or newcomers from other industries with no proven robotics experience.

The result is a landscape flooded with technologically almost identical prototypes that can fold shirts and wave in perfectly lit demonstrations, but require human intervention after no more than 40 minutes in real-world production environments. While the Ministry of Industry and Information Technology has set ambitious goals—mass production by 2025, fully autonomous factory operations by 2030—the 15th Five-Year Plan envisions the true commercialization of humanoids only toward the end of the planning period. The gap between political narrative and commercial reality is the real stress test the industry is currently facing.

This situation can be described as the "valley of death": the critical phase between technological maturity and economic viability, in which companies can no longer survive on research funding alone, but not yet on market revenue. For Chinese humanoid startups, this valley has a very concrete dimension: the market is estimated to reach around 10.47 billion yuan (approximately US$1.4 billion) by 2026 – which sounds impressive, but is divided among more than 150 competitors. The successful companies have therefore begun to undergo a strategic paradigm shift: away from waiting for the perfect robot, towards generating immediate cash flow with what is available today.

From the stage to a fixed operating asset – entertainment as the primary cash flow engine

The most obvious interim solution was the entertainment industry, but the transition from a short-term attraction to a sustainable business model required a significant shift in entrepreneurial thinking. In the initial phase, humanoid robots were primarily in demand as one-off attractions at trade fairs, galas, and opening ceremonies – the novelty effect quickly faded, and the cost-benefit ratio remained unsatisfactory. The decisive change came with productization: robots were no longer treated as technological demonstrators, but as manageable assets that could be integrated into existing commercial structures in the long term.

The figures impressively demonstrate this transformation. China's robot rental market grew from approximately 140 million yuan to over one billion yuan in a single year – a tenfold increase in twelve months. Companies like AgiBot launched dedicated leasing platforms: AgiBot operates "Qingtian Rent," a nationwide service already active in 50 cities, connecting over 1,000 robots and 600 service providers. Unitree Robotics and AgiBot report fully booked calendars for corporate events, weddings, and trade shows, with daily rates ranging from 200 yuan for basic robot dogs to 10,000 yuan for sophisticated interactive humanoids. Furthermore, the launch of BOTSHARE, China's first open robot leasing platform, created a marketplace-like infrastructure that further scales the model.

The real economic advantage of this approach, however, lies deeper than direct rental income. First, entertainment requires significantly lower technological reliability thresholds compared to industrial production: A robot that occasionally makes a mistake at a gala is charming; the same mistake on an automotive assembly line costs efficiency and quality. Second, short-term rentals—typically one to three days—generate extensive usage data under real-world conditions, which is valuable for the further development of control algorithms. Third, the model operates according to the logic of an infrastructure investment: Instead of tying up capital in robots gathering dust in warehouses, revenue flows in from simple use cases while the actual target applications are still being developed. The transition from event robotics to permanent integration in theme parks, museums, and exhibition halls with their own IP shows marks the next stage of maturity: No longer temporary bookings, but multi-year contracts with ticket revenue sharing or fixed lease payments create the stable cash flow that investors demand.

Data as a strategic resource – the development of “Physical AI” ecosystems

Anyone who has followed the global race for artificial intelligence knows the basic principle: whoever controls the most and highest-quality training data wins the next stage of AI development. In the field of language models, this was the internet. In the field of physical AI—that is, AI that controls real bodies in the real world—it is high-quality motion and interaction data from teleoperated robots. This realization has led to the emergence of an entirely new industry in China: the construction of state-funded data collection centers as a commercially viable infrastructure.

The scale of this strategy is remarkable. China has established over 40 specialized robot training centers where human operators, equipped with VR headsets and exoskeletons, guide humanoid robots through everyday tasks, recording every movement as training data. The largest of these centers, built in Beijing's Shijingshan district in partnership with Leju Robotics, covers over 10,000 square meters and features 16 different training scenarios. The center in Zigong, Sichuan, is designed to cover 6,000 square meters and, at full capacity, is expected to produce 15,000 data sets daily—three million high-quality entries annually. Cities from Beijing to Shanghai, Zhengzhou, and Zigong are competing to host this infrastructure, much like cities once vied for semiconductor factories.

The monetization logic of this model is multifaceted. At the first level, robot manufacturers sell their machines directly to these data centers: UBTech alone generated 566 million yuan (around US$80 million) in revenue from sales to three such centers in Jiangxi, Guangxi, and Sichuan. China Mobile placed orders totaling 124 million yuan (US$17.6 million). At the second level, the datasets generated in this way become a tradable commodity: standardized "motion corpora" that serve not only the data-collecting companies themselves, but also external AI developers and technology companies as a training basis for robot control models. The business model thus shifts from pure hardware production to a hybrid approach, in which the hardware serves to generate data, and the data, in turn, is marketed as a scalable product.

A structurally crucial factor is the integration of vocational schools and universities as suppliers of qualified yet cost-effective teleoperators. In a country with structurally high youth unemployment—most recently exceeding 18 percent among urban youth—this creates employment opportunities that are socially beneficial to the state and economically advantageous to businesses. An entirely new profession has emerged: the AI ​​robot trainer, who is part choreographer, part data scientist, and part instructor. Training a single, simple grasping device requires tens of thousands of repeated movements; a single data collection session can cost over 1,000 yuan—which significantly underscores the value of high-quality, standardized datasets. Those who build a scalable data infrastructure today create a structural competitive advantage that can be measured in years of data superiority.

High risk instead of high volume – niche markets as an economic bridge

The third survival strategy addresses a fundamental contradiction in the humanoid robotics business model: The most obvious market – the mass production line – is still too demanding for the current generation of technology, while seemingly smaller markets are a significantly better fit for the capabilities of today's robots. Automotive production requires cycle times in the range of seconds and near-zero error rates; the current state of the art typically allows 20 to 40 minutes of continuous operation without human intervention. For high-risk inspections in power grids, however, the primary concern is that a robot is safer than a human – not whether it operates at industrial speed.

This principle has already led to large-scale implementations in the Chinese energy sector, which can serve as a blueprint for the entire industry. China's state-owned grid operator, State Grid, announced the procurement of 8,500 embodied AI robots with a total budget of 6.8 billion yuan (approximately one billion US dollars). Particularly noteworthy is the high-voltage live-line operations segment: 500 humanoid robots are being acquired with a budget of 2.5 billion yuan (370 million US dollars) to replace humans in high-risk work on distribution networks and ultra-high-voltage projects. The economics of this decision are compelling: According to State Grid, each deployed AI unit is expected to save between 500,000 and 800,000 yuan (70,000 to 110,000 US dollars) in annual labor costs, with a payback period of approximately two to three years. Inspection efficiency increases fivefold, error response time decreases by 60 percent, and over 90 percent of human exposure to high-risk operations is eliminated.

These figures illustrate why high-risk niche markets are more economically attractive than directly targeting mass markets: Willingness to pay is exceptionally high because the value of risk reduction is immediately measurable for clients. Mining, chemical plants, nuclear facilities, and disaster relief follow the same logic. Because safety is prioritized over speed, the technological hurdle is significantly lower compared to automotive manufacturing – and margins are considerably higher. The model is thus one of targeted market positioning: not fighting against the strongest competitor, but rather first developing the most accessible, highest-paying market.

In the education sector, Chinese manufacturers are pursuing a parallel, classic "shovel-seller" model: universities, vocational schools, and research institutions are purchasing humanoid robots for teaching and laboratory use. UBTech, for example, has integrated its Walker models into training programs offered by automakers such as BYD, NIO, and Geely. This market offers more reliable cash flows than industrial pilot projects because education budgets are less dependent on economic cycles and institutional buyers have long procurement cycles. Unitree Robotics has also launched the G1, a humanoid robot explicitly designed for research and education, at a price of 39,900 yuan (around US$5,500)—a clear indication that a volume strategy in the education market is understood as the foundation of its large-scale model.

Sino-Cooperation is a platform based in China and Germany

Sino-Cooperation is a platform based in China and Germany that promotes exchange and cooperation between German and Chinese companies, especially through events, digital formats and an online cooperation exchange for market entry and partnerships.

More information here:

• Sino-cooperation

Who survives – an empirical inventory of the first winners

Behind these strategic considerations lies the question of which companies are already providing initial, reliable evidence that these survival strategies are working. UBTech is currently considered a clear benchmark. The Shenzhen-based company, regarded as China's first publicly listed humanoid robot company, reported total revenue of 2.001 billion yuan for 2025 – a 53.3 percent increase year-over-year. The most spectacular single figure: Revenue from full-size embodied AI humanoids jumped from 35.6 million yuan in 2024 to 821 million yuan in 2025, a rise of 2,203.7 percent. Cumulative sales of the Walker S-series reached 1,079 units – a 35,866 percent increase year-over-year, although this figure reflects the extremely low starting point. The gross margin increased from 28.7 percent to 37.7 percent. The net loss decreased by 31.9 percent to 790 million yuan.

These figures illustrate both the progress and the remaining weaknesses of the sector. Despite impressive revenue growth, UBTech remains unprofitable, and trade receivables rose to 1.302 billion yuan—partly due to late payments from government clients, indicating structural risks in converting orders into cash. The company is responding with vertical integration: The acquisition of a 29.99 percent stake in servo motor manufacturer Zhejiang Fenglong Electric secures strategic supplier relationships and reduces dependencies.

Noetix Robotics, a younger company, illustrates the investment logic: Founder Jiang Zheyuan emphasized after a pre-B funding round of over 300 million yuan that humanoids still need to "find new use cases" and that expansion into the right niches will be crucial in the face of increasingly intense competition. This isn't marketing rhetoric, but a clear description of the imperative for survival: Companies that occupy the right niche early and accumulate data and operational experience there build advantages that are structurally difficult to overcome. Unitree, which addresses a broader market base with its cost-effective H1 and G1 models, is rapidly gaining market share, particularly in research and education segments—a model reminiscent of Tesla's early strategy: using a lower-cost, high-volume model to build the infrastructure for higher-value segments.

The state ecosystem as a guarantor of survival – and a structural risk

No picture of Chinese humanoid robotics would be complete without a critical assessment of the state support architecture that enables the survival of many companies. The state operates on several levels simultaneously: as a promoter through direct subsidies and tax breaks (local governments have provided approximately 120 billion yuan to date), as a provider of space by offering free or heavily subsidized areas for offices and production facilities, as an early adopter by procuring from government agencies and state-owned enterprises like State Grid, and as a regulator by establishing national standardization systems.

The mechanism of state-sponsored early adoption is particularly effective. When government agencies and state-owned enterprises procure humanoid robots for museum tours, traffic monitoring, or industrial inspections, they effectively cover the costs of real-world testing under production conditions—a privilege not enjoyed by private Western competitors. In China's burgeoning robotics hubs, such as the "Robot Valley" in Shenzhen, billions are being invested in the development of AI models and robot hardware. Furthermore, buyers sometimes receive a partial refund of up to ten percent of the purchase price, further lowering the barrier to acquisition.

This support architecture is both the sector's strength and its vulnerability. The strength is obvious: Chinese startups can iterate and fail without being immediately penalized by the market, which significantly accelerates technological learning curves. The vulnerability lies in the distortion of market signals: When government procurement and subsidies generate the majority of demand, companies develop products for government requirements rather than genuine market needs. The NDRC's warning about bubbles is, not least, an admission that its own support architecture has led to an overproduction of homogeneous, unmarketable products. Furthermore, delayed payments from government clients to UBTech signal that even government demand is not a secure basis for short-term liquidity.

China versus the rest of the world – structural asymmetries in global competition

A comparison between Chinese and Western approaches to humanoid robotics reveals fundamental differences in economic logic, not just in technology. American companies like Boston Dynamics, Figure AI, and Tesla Optimus rely on proprietary technology, high-priced early adopters, and the appeal of their capital market history—the classic Silicon Valley approach. Chinese companies, on the other hand, pursue a strategy reminiscent of the successful industrialization of electric vehicles: mass production, cost reduction, government support for early market penetration, and aggressive economies of scale.

The decisive structural advantage of Chinese manufacturers lies in their local supply chain. Critical components such as sensors, batteries, servo motors, and actuators can be sourced almost entirely domestically, minimizing response times and significantly reducing development costs. Sixty-one percent of all humanoid robot concepts worldwide since 2022 originated in China—an indicator of both technological breadth and the sheer number of active developers. China's share of the global humanoid market exceeded 80 percent of worldwide installations in 2025, with global revenue surpassing US$500 million for the first time. The Chinese company UBTech ranks first among global humanoid robot companies in terms of revenue from full-size humanoids.

For Europe and Germany, this assessment reveals an uncomfortable truth: the window for establishing their own market positions in the basic infrastructure of humanoid robotics is closing. In its April 2026 analysis of China's strategy for embodied AI, the Merics Institute noted that the country is using its industrial robotics base as a springboard, and despite impressive public demonstrations, its real-world capabilities in actual production scenarios are still limited to pilot and demonstration projects. The International Federation of Robotics (IFR) realistically assesses that the widespread adoption of humanoids as universal factory assistants is not in the near or medium term – underscoring the relevance of the described strategic shift.

The economics of transition – what will shape the industry in the long term

The true merit of the three survival strategies described lies not solely in short-term cash flow generation, but in their impact on medium-term competitiveness. Entertainment and leasing generate real operational data and brand presence. Data collection centers create strategic resources for training the next generation of models. Niche applications in high-risk sectors create reference projects that facilitate entry into larger markets. All three approaches share a common principle: They fully exploit today's technological capabilities instead of waiting for future ones.

This pragmatism offers a tangible economic advantage over the narrative approach—the model primarily based on investor stories and postponing commercial reality into the future. Companies that find paying customers for interim solutions today are accumulating three types of capital simultaneously: financial capital (cash flow and reduced capital dependency), technological capital (real-world operational data for AI training), and institutional capital (customer relationships, reference projects, market knowledge). Companies that rely primarily on funding rounds do not possess any of these three types of capital to a substantial degree.

The market consolidation that the NDRC actively seeks to promote will most likely follow this dividing line: Companies with robust revenue streams—whether through leasing, data sales, or niche applications—will survive the consolidation pressure. Companies relying solely on demonstrative technology and investor funding will be acquired or go bankrupt in a wave of consolidation. This is ultimately just the well-known logic of industrial maturation processes, applied to an industry that is still in an exceptionally early stage. The pace of this maturation—and thus the pace of market consolidation—will depend significantly on how quickly leading companies can translate the quality of their real-world operational data into superior AI models that raise autonomy and reliability to an industrially viable level.

Technological understanding is a must – business acumen is the real bonus

The described development in the Chinese humanoid robotics sector is not an isolated industry phenomenon, but rather exemplifies a broader pattern of Chinese technology industrialization: aggressive government support in the early stages, followed by pragmatic market structuring once the bubble becomes apparent. The three identified survival strategies—entertainment and leasing, data production, and niche applications—are not the result of central planning, but rather of adaptive entrepreneurial responses to the same economic constraints faced by every startup sector.

What makes the Chinese case particularly striking is the intersection of state infrastructure and private-sector pragmatism. Publicly funded data collection centers, state-sponsored early procurement, and subsidized suppliers create an enabling framework that structurally disadvantages Western competitors. Within this framework, however, the same fundamental entrepreneurial principles that apply in any other market ultimately determine survival and failure: those who acquire paying customers early, systematically accumulate data, and build relationships with high-value customers secure a starting point for the actual industrial breakthrough. This breakthrough—considering all technological and regulatory developments—likely lies in the second half of this decade. The companies that have crossed the valley of death by then will likely be in a position that is structurally almost impossible to overcome.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your native language!

Konrad Wolfenstein

I and my team are happy to be available to you as your personal advisor.

You can contact me by filling out the contact form here simply call me at +49 7348 4088 965. My email address is [email protected]:or

I'm looking forward to our joint project.

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs

When proven strategies fail: Organizational adaptability in the digital transformation of ambidexterity - Image: Xpert.Digital

We are currently experiencing a period of economic turmoil that differs fundamentally from previous recessions. A deceptive silence prevails in the boardrooms of European and international companies – broken only by the sound of failing strategies that were considered a guarantee of success just yesterday. This is not merely a cyclical downturn, but a profound structural break. The tools with which companies achieved growth for over two decades simply no longer work.

More information here:

• When proven strategies fail: Organizational adaptability in the digital transformation of ambidexterity