China's humanoid robot sector attracted more than 400 billion yuan ($56 billion) in funding during 2025, yet total industry shipments barely reached 10,000 to 20,000 units, exposing a widening gap between capital enthusiasm and commercial viability that threatens to reshape the nascent industry.

The stark contrast between investment inflows and actual deployments has triggered the first wave of high-profile failures. Yixing Robotics, backed by Geely Group, dissolved within five months of securing hundreds of millions in yuan funding. CloudMinds, once valued at over 20 billion yuan (US$2.8 billion), reportedly abandoned its headquarters as suppliers pursued debt collection. Silicon Valley's K-Scale shut down less than two years after seed funding, refunding pre-orders as cash reserves dwindled to US$400,000.

Industry insiders confirm that leading domestic manufacturers shipped only thousands of units each in 2025, with mid-tier companies delivering hundreds. This compares unfavorably with China's 295,000 industrial robot installations in 2024, underscoring humanoid robots' struggle to transition from demonstration projects to production-scale deployments.

The funding frenzy has created what analysts describe as a "performance-driven" cycle, where capital flows primarily support technology demonstrations rather than revenue-generating applications. Most purchasers remain research institutions and universities, while factory orders—the critical commercial validation—have largely failed to materialize at scale.

Capital Flood Masks Commercial Drought

The 162 financing rounds completed through November 2025 represented a threefold increase over 2024 levels, according to Qianzhan Industry Research Institute. Early-stage investments dominated, with angel and Series A rounds comprising over 70% of deals, indicating most projects remain in "scientific commercialization" phases with unproven business models despite elevated valuations.

Robotics Intelligent Technology secured over US$120 million across two angel rounds within months of establishment. Galaxy General raised US$300 million in a single round, setting an industry record. Leju Robotics closed a 1.5 billion yuan (US$210 million) pre-IPO round, while Unitree Technology completed IPO counseling in just 132 days, the fastest among technology companies this year.

Secondary markets mirrored the enthusiasm. The A-share humanoid robot concept sector gained 54.98% through mid-December 2025, significantly outperforming broader indices. Companies merely mentioning humanoid robots in announcements frequently triggered next-day price movements, prompting one securities analyst to privately observe that "the A-share market needs humanoid robots more than humanoid robots need the A-share market."

Yet capital flows diverged sharply from product flows. Unitree Technology disclosed that four-legged robots contributed approximately 65% of 2024 revenue, with humanoid robots accounting for just 30%—almost entirely sold to universities and research institutes for educational and research purposes, with zero industrial deployments.

Order Book Complexity

Second-half 2025 saw headline-grabbing order announcements that initially appeared to validate commercialization momentum. UBTech Robotics reported cumulative orders exceeding 800 million yuan (112million)foritsWalkerseries,withtotalannualorderssurpassing1.3billionyuan(112million)foritsWalkerseries,withtotalannualorderssurpassing1.3billionyuan(182 million). Hundreds of full-size Walker S2 industrial robots entered production and began deployment at BYD, Geely, Foxconn, and SF Express facilities.

Yuejiang signed an 80.5 million yuan (US$11.3 million) procurement contract with precision manufacturer Ruidefeng. Zhiyuan Robotics and Unitree jointly won a 124 million yuan (US$17.4 million) tender. Reports circulated that Tesla issued US$685 million in linear actuator purchase intentions to Chinese suppliers including Sanhua Intelligent Controls.

However, Morgan Stanley's year-end analysis revealed that a substantial portion of these "large orders" constituted framework agreements or letters of intent rather than binding, irrevocable purchase contracts, leaving execution uncertainty. Industry sources indicated some orders involved related-party transactions or government-coordinated "demonstration projects" funded by local industrial guidance funds, typically requiring three-to-six-month operational validation with partial payment upon acceptance, but lacking clear plans for subsequent reorders or scaled replication.

Goldman Sachs' November field research across nine Chinese humanoid robot supply chain companies provided corroborating evidence. While most suppliers planned capacity for 100,000 to one million equivalent robots annually, none confirmed receiving large-scale orders or definitive production schedules. Goldman's own forecast projects global humanoid robot shipments reaching just 1.38 million units by 2035.

"The industry must distinguish whether current production stems from genuine commercial demand or pseudo-demand created solely by policy subsidies and investment fervor," stated Wang Zhongyuan, director of the Beijing Academy of Artificial Intelligence. "If driven only by one-time demonstration project requirements, procurement will halt once implementation results disappoint, potentially plunging the sector into a trough."

Technical Bottlenecks Beyond Motion Control

Motion control, once considered the primary technical challenge, actually showed the fastest progress in 2025. Robots successfully performed complete dance routines on stages, ran half-marathons autonomously, and executed dodges and counterattacks in boxing matches. Industry participants assess that domestic companies achieved "85-point" performance in hardware and fundamental motion control with competitive cost structures.

The genuine bottleneck has shifted to the "brain"—perception, cognition, and decision-making systems. Wang Xingxing, founder of Unitree Technology, articulated the problem directly: Vision-Language-Action (VLA) models encountered significant challenges in 2025. These models, trained on internet text and image data, perform impressively in conversational applications but struggle with the continuous, dynamic, three-dimensional or four-dimensional spatiotemporal trajectory data that robots require. Physical world scarcity and complexity have emerged as core constraints on model capabilities.

A typical scenario illustrates the challenge: instructing a robot to "pick up that water bottle from the table" requires three-dimensional spatial positioning to "see" the bottle—yet most robots remain limited to two-dimensional image recognition, unable to accurately assess wall locations or target distances. Planning the grasp demands understanding the bottle's material, weight, and cap status to determine finger count and force application. Executing the movement involves force control and tactile feedback, with outcomes varying based on whether the bottle is inverted, full, or half-empty.

Zhang Hengdi, co-founder of Pacinian Perception Technology, offered an analogy: robots without tactile sensing represent "information-deficient" systems. Tactile integration transforms robots from "precisely pre-programmed executors" into "explorers capable of adjustment and adaptation."

Diverging Paths and Consolidation Pressures

Not all companies will survive until the anticipated "ChatGPT moment" for humanoid robots arrives. CloudMinds' multiple regional offices were reportedly abandoned in spring 2025. The company, which raised over 5.4 billion yuan (US$756 million) cumulatively with peak valuations exceeding 20 billion yuan (US$2.8 billion), suffered capital chain rupture after two failed listing attempts, leading to office space rentals, employee wage disputes, and supplier debt collection.

Yixing Robotics, established in May 2025 with backing from Geely Group and actual control by Li Xingxing (son of Li Shufu), completed hundreds of millions in yuan financing within one month of founding. By October, the company entered dissolution proceedings after its business direction overlapped with Geely's priority support for Qianli Technology, prompting complete withdrawal of group-affiliated personnel and eliminating core resource support within five months.

Despite mounting failures, China became the country with the most humanoid robot startups globally. Statista data shows 23 China-based humanoid robot-focused startups by end-2025, narrowly ahead of the United States' 22. Global humanoid robot startup funding reached $2.65 billion in 2025, exceeding the combined total from 2018 through 2024 for the first time.

This apparent contradiction reflects standard narratives in emerging industries approaching maturity: bubbles burst, but capital consolidates toward companies with genuine technical barriers and scenario understanding rather than exiting entirely. Unitree Technology and Zhiyuan Robotics both exceeded 5,000 units in 2025 production, becoming the only two manufacturers globally achieving scaled shipments. UBTech initiated mass production delivery of hundreds of industrial robots in Shenzhen.

Accounting Question Awaits Answer

The Ministry of Industry and Information Technology formally established the Standardization Technical Committee for Humanoid Robots and Embodied Intelligence in December 2025, signaling unified standards to prevent fragmented development and redundant efforts.

The same month, Goldman Sachs updated research on supply chain companies including Sanhua Intelligent Controls and Tuopu Group, noting that while most suppliers planned extremely aggressive capacity—100,000 to one million equivalent robots annually—none confirmed receiving large-scale orders or clear production timelines.

These developments encapsulate the humanoid robot industry's tension at 2025's close: universal conviction that the future will arrive, paired with complete uncertainty about timing and form. The sector completed its transformation from laboratory curiosity to capital market protagonist, ascending mainstream stages and receiving enthusiastic reception while experiencing its first brutal shakeout.

Yet it has not answered the most fundamental question: What cost savings or revenue generation does purchasing a robot actually deliver? Is the investment payback period 36 weeks, 36 months, or incalculable?

Companies promising delivery this year and profitability next during fundraising now face their "homework submission" moment. Framework agreements filled with optimistic projections will undergo financial audit scrutiny. Robots operating successfully in demonstration projects must maintain three consecutive months of fault-free operation without engineer supervision.

The path extends longer than anticipated. But 2025 confirmed one critical fact: humanoid robots have entered a technical convergence phase where engineering problems can be systematically decomposed and sequentially solved, no longer confined to science fiction imagery.

Capital deployed 40 billion yuan ($5.6 billion) for a platform ticket to the future. Now, boarding time approaches.