Robots Born in 2025 Are Already Facing a Bottleneck

The Bubble Is Taking Shape

Signs of froth are becoming increasingly hard to ignore. Over the past six months, multiple investors and founders active in the embodied intelligence space have told Huxiu in unambiguous terms: most embodied robot manufacturers are facing an acute “order drought.” Across both B2B and consumer markets, real demand may no longer be sufficient to absorb the volume of robots scheduled for production in 2025.

A deeper structural risk is also emerging. Many high-profile embodied intelligence startups that secured funding in 2025—some of which have already grown into mid-tier industry players—are now trapped in a business model dilemma. With limited orders, these companies lack the ability to generate healthy, self-sustaining cash flow.

One well-known investor put it bluntly: “Right now, in embodied intelligence, the real buyers are investment institutions and the internet giants they invest in. In simple terms, there is no true ToB or ToC—this is essentially ToLP (to limited partners).” In his view, aside from a handful of top players such as Unitree and Zhiyuan that have begun to explore relatively mature commercialization paths, most companies’ survival logic for 2026 remains centered on “convincing capital to stay in,” rather than genuinely cracking B2B or B2C models.

A Rush of Dreamers into the Same Lane

Yet the other side of the coin is equally striking. Despite mounting concerns, waves of ambitious entrepreneurs continue to pour into embodied intelligence.

Around the time of that conversation, I met a serial entrepreneur who had spent over two decades in the traditional automotive supply chain. He had worked on complete vehicles, key components, and nearly every corner of the industry. Before we met, he told me he was about to begin a brand-new journey—his final startup. As soon as he sat down, he said with excitement and urgency, “I’m going to build robots. Embodied intelligence.” The tone made it feel as though missing this window would mean missing everything.

The scene felt familiar. Over the past few months, I’ve encountered dozens of similar cross-industry founders—many from automotive, autonomous driving, industrial equipment, or even unrelated traditional sectors—who spoke with the same intensity. Almost all mentioned Elon Musk. Almost all were convinced this was the right direction. And almost all radiated a strong sense of urgency.

This has become a defining snapshot of the embodied intelligence world over the past year. More and more entrepreneurs are redirecting their focus toward embodied intelligence. Some jump in directly, others stake early claims through investment or M&A, and listed companies increasingly highlight “embodied intelligence” and “humanoid robots” in announcements, roadshows, and external communications.

Calm investors and fervent entrepreneurs now form a stark contrast within the industry. On one side are immature products and unresolved business models; on the other, unprecedented enthusiasm to enter the race. More players, using different paths, are squeezing into the same track.

Against this backdrop, an unavoidable question emerges: beneath this seemingly surging wave, what is the industry’s true condition?

Unproven Models: Inflated Research Demand, Misaligned Industrial Reality

From a surface-level perspective, embodied intelligence in 2025 has almost become a “blind pick” option in the startup world. Capital markets have amplified this signal—multi-billion-yuan funding rounds, record-breaking single investments, and frequent claims of “industry firsts” and “domestic breakthroughs.” Add to that robots dancing, serving tea, and performing on short-video platforms and variety shows, and a still-nascent sector has been thrust into the spotlight.

But once the stage lights fade, the reality of deployment looks far more limited.

Within the industry, the most frequently cited application scenario remains research. Huxiu learned that for nearly all leading embodied intelligence companies, the majority of orders still come from research institutions. Data suggests that total industry orders in 2025 exceeded RMB 90 billion, with research-related demand growing rapidly.

Yet an unspoken consensus exists inside the industry: research scenarios are inherently transitional.

Whether it’s robots frequently appearing in universities, laboratories, and exhibitions, or the research and demonstration projects undertaken by leading manufacturers, these scenarios resemble a short-term carnival. They validate technology and sustain narratives, but from the outset, they lack the potential for scalable replication.

Multiple industry insiders told Huxiu that robots purchased by universities are often not full humanoid systems, but rather single robotic arms, limb-less bodies, or partial modules used solely for algorithm research. These are research tools, not production-ready products.

As a result, this demand is difficult to translate into real productivity, and its procurement scale is inherently limited.

Industry Demand Exists—but Supply Misses the Mark

Industry consensus still holds that industrial scenarios are the most promising path for real deployment. The challenge, however, is not a lack of demand, but a clear mismatch between current robot capabilities and factories’ actual needs.

Practitioners note that before humanoid robots can enter factories, they often require long commissioning cycles. Initial debugging alone typically takes seven to eight months, and even under ideal conditions, reaching a demonstrable prototype level takes three to four months. This pace directly conflicts with factories’ requirements for stability, certainty, and return on investment.

From the demand side, factories do not lack imagination; they lack certainty.

Huxiu learned through discussions with multiple manufacturers that when factoring in purchase costs, deployment and tuning expenses, and ongoing maintenance, the total cost per robot often exceeds human labor costs. This significantly raises the threshold for large-scale adoption in industrial settings.

Upstream robotics company Annu Intelligence believes that realistic deployment opportunities remain highly concentrated in industrial verticals, which is why its own product portfolio focuses exclusively on industrial domain models. This assessment reflects the current state of commercialization.

Even so, Annu Intelligence CEO Wen Hongjie told Huxiu that actual industry sales this year remain at only the low thousands of units—still far from the “tens of thousands” scale.

There has been tangible progress. For example, Fulin Precision has deployed around 80 robots and completed its proof-of-concept phase. Annu Intelligence also revealed that a major chip company will sign a cooperation agreement next month, with expected deployment reaching triple-digit volumes.

Yet such cases remain exceptions. Only two or three top-tier companies are securing meaningful industrial orders, and many financing events still involve strategic investments from industrial shareholders, with a significant portion of orders tied to related-party transactions. These deals do not yet represent a fully market-driven demand explosion.

Within this reality, industry insiders view the “bubble” with more restraint than outsiders might expect.

In Wen Hongjie’s view, the real bubble does not stem from immature technology itself, but from three types of players: those rushing in to chase hype, those unclear about application scenarios, and those making excessive early-stage promises to customers.

Short-Term Cash Flow Over Long-Term Answers

Before robots can become large-scale productive tools, structural tensions are forcing companies to seek immediate commercialization outlets. This explains why robot leasing and performance businesses surged in 2025.

One example is Qingtian Lease, a robot leasing platform backed by Zhiyuan Robotics. Its CEO Li Yiyan told Huxiu that the company does not bind itself to any single manufacturer. Meanwhile, Zhiyuan partner Jiang Qingsong sees strong growth potential in leasing, noting that the market generated roughly RMB 10 billion in revenue in 2025 and could exceed RMB 100 billion in 2026.

With large-scale industrial deployment still out of reach, robot performances and short-term rentals have become the fastest ways to generate cash flow and visibility. They bypass complex production environments and provide a compromise path for companies committed to training models with real data but constrained by limited real-world scenarios.

From a commercialization standpoint, however, these are transitional stopgaps, not sustainable answers.

As Jiang Qingsong acknowledged, before robots truly enter households, entertainment-focused commercialization is inherently a temporary phase.

Highly anticipated eldercare scenarios remain distant. Even seemingly simple tasks like pouring water or handing items to seniors may still require three to five years to become viable. Truly productive industrial scenarios may take five to ten years to reach maturity.

Dexterous Hands: The Hottest—and Bubbliest—Component

Another lens on the embodied intelligence bubble lies in its most expensive hardware component: dexterous hands.

Dexterous hands, which replicate human hand functions with high degrees of freedom, are critical actuators for fine interaction. According to QYResearch, the global market entered sustained expansion in 2024, reaching USD 815 million and projected to grow to USD 10.3 billion by 2031, with a CAGR of 40.4%.

By the end of 2025, the sector had seen over 55 financing events, with disclosed funding exceeding RMB 7.5 billion. Lingxin Qiaoshou alone completed five rounds of financing in nine months.

As humanoid robot hype intensifies, demand has spilled over into core components. Guangyuan Capital vice president Quan Bo attributes the surge to three factors: policy support following the Ministry of Industry and Information Technology’s humanoid robot guidance, demand pull from robot manufacturers, and capital overflow from industry leaders into upstream components. Overall, he believes the market remains relatively rational.

Some investors expect 2026 funding to shift further upstream.

Guangyuan evaluates dexterous hand projects based on technical depth, productization and cost control, and ultimate commercialization potential. The long-standing challenge remains the “impossible triangle” of performance, cost, and reliability.

Technical paths are still diverse, and different scenarios will likely support different solutions. Over time, Quan Bo believes large-scale commercialization and supply chain maturation will reinforce each other, driving cost reductions and stability.

Prices currently vary widely—from tens of thousands of dollars abroad to sub-RMB-10,000 offerings domestically. While it’s difficult to quantify future cost reductions, Quan Bo believes prices will eventually reach commercially viable levels.

Where the Bubble Truly Comes From

In 2025, the robotics industry has already seen its first real wave of retreat. Several companies across consumer robotics, companion robots, collaborative robots, and warehouse automation have shut down, declared bankruptcy, or quietly ceased operations—including long-established names like iRobot.

Ironically, these exits occurred at the industry’s most inflated moment. Hundreds of billions of yuan poured in, creating countless highlight moments.

Domestically, embodied intelligence companies have clearly stratified into tiers. Top-tier players—such as Unitree, Zhiyuan, Galaxy General, Beijing Humanoid Innovation Center, Yunshen Chu, and Fourier—enjoy the most attention and funding.

Yet even within this group, operational realities diverge sharply. The highest revenue player approaches RMB 20 billion annually, while others hover around RMB 5 billion. Many still rely heavily on project-based, showcase-oriented, or related-party revenues, leaving business model stability unproven.

At the macro level, the mismatch is even starker. From January to October 2025, primary market funding exceeded RMB 500 billion, while total annual revenue across all top-tier companies remained below RMB 100 billion. Capital and income have diverged by multiples.

This gap is the core source of the bubble.

Much like the large-model boom two years ago, the root cause lies in valuation mismatch. Many investors are driven by FOMO, applying large-model investment logic to embodied intelligence. But humanoid robots follow manufacturing logic, not software logic. Industrial adoption demands long cycles, strict validation, and low tolerance for error—often five years or more.

Today’s capital structure does not align with this reality. Valuations are pulled forward, but timelines are not.

This misalignment is not unique to China. In the U.S., bubbles in embodied intelligence may be even larger, with startups raising massive funding without clear orders. Yet long-term optimism still centers on players like Elon Musk, whose advantage lies not in concepts, but in access to scalable real-world demand.

The same logic applies domestically. Companies with manufacturing capacity, application scenarios, and organizational strength hold true long-term advantages—not those surviving solely on narratives.

Under these conditions, IPO pressure is building. Multiple companies are rumored to be planning listings, prompting regulators to warn against clustered listings and compressed R&D space.

As one industry executive told Huxiu, embodied intelligence remains in an early exploratory phase, and now is not the right time for mass listings. But once valuations rise and exits are tied to capital cycles, IPOs become less a choice than an inevitability.

It is within this growing misalignment that the bubble continues to inflate.