Embodied AI Funding Nears 100 Billion Yuan, But Fewer Than 20% of Robots Reach Factories
In the first half of 2026, China recorded 172 embodied AI financing deals totaling 109.174 billion yuan. The number of embodied AI companies valued at over 10 billion yuan rose from 3 in 2025 to 25 by August 2026; some dexterous hand makers saw valuations jump from under 1 billion to 20 billion yuan in less than two years.
An iiMedia report notes that embodied AI is in a critical accumulation phase, moving from proof-of-concept to engineering deployment, with 2026 as the inflection point. The core contradiction has shifted from “can it be built” to “can it be used stably, scaled, and made profitable.”
The money is flowing in, but fewer than 20% of robots actually enter factories. In H1 2026, global humanoid robot shipments exceeded 22,000 units, up nearly 300% year-on-year; over 60% went to entertainment and education/data collection, while less than 20% entered smart manufacturing and warehouse logistics.
An investor noted that a million-yuan order can leverage hundred-billion-yuan valuation imaginations for upstream suppliers, creating a widening rift between capital narratives and industrial reality. The core difficulty is the significant gap between “can demo” and “can work.” At the 2026 JD Global Tech Explorer conference, industry figures said about 70% of products require deep reconstruction to meet real scenario needs. The root cause: early R&D followed academic logic—payload, end-effectors, and whether dexterous hands are needed were not reverse-engineered from actual operational demands.
In August 2026, Unitree and UBTech reported interim results. Both humanoid robot businesses hit record highs in H1, with revenues at the same scale: UBTech 1.269 billion yuan, Unitree 1.152 billion yuan—but different business logic.
Unitree’s H1 2026 revenue was 1.152 billion yuan, up 48.54% YoY, with net profit attributable to shareholders of 274 million yuan, turning profitable. However, non-GAAP net profit actually fell 19.34%, meaning real profitability did not improve. In Jan-Sep 2025, 73.60% of its humanoid robot revenue came from research institutes and tech firms, with industrial scenarios accounting for only about 2.60%. Its profit largely depends on novelty purchases from the research market, not true industrial scale.
UBTech’s H1 2026 revenue was 1.269 billion yuan, up 104.2% YoY, with gross margin rising from 31.50% in 2023 to 44.70%; clients include Airbus, BYD, and Geely. But it still lost 339 million yuan in the period, though the loss narrowed 23% YoY—not yet breakeven. R&D expenses were 303 million yuan, a 23.9% R&D expense ratio. H1 revenue included 139 million yuan from the consolidation of Fenglong Shares, a non-robot business; excluding that, original robot business revenue was about 1.13 billion yuan, slightly lower than Unitree. Meanwhile, traditional education robot revenue fell 49.1% YoY, its share plunging from 48.7% to 12.7%.
The report states that embodied AI commercialization strictly follows a gradient: specialized scenarios first, general forms later. Short-term growth continues to come from B-end scenarios with highly structured workflows and clear labor replacement value. Zhang Yi, CEO and chief analyst of iiMedia, noted that the biggest shortcoming is not whole-machine R&D, but that once mass production begins, support systems for delivery, maintenance, data collection, and model training are almost blank. The industry singularity is expected around 2030; before then, deepening specialized scenarios and accumulating technology and data are core tasks. Unitree’s profit-quality dilemma and UBTech’s scale loss-reduction curve point to one fact: in embodied AI, the survivors won’t be the most funded or highest valued, but those who first learn to do the math.