Shenwan Hongyuan Group Co., Ltd. released a research report stating that 2026 is the pivotal first year for Physical AI to break away from Screen AI. The robotics industry is modeled as a hybrid of "smartphones + autonomous driving," and investment should follow the new paradigm of "Intelligence Layer > Coordination Layer > Hardware Layer," focusing on core capabilities and ecosystem building. The firm's primary focus is on companies with data closed-loop and scenario capabilities; followed by enterprises with data scenarios and related expertise in the metaverse; and finally, high-quality component manufacturers. Shenwan Hongyuan's main views are as follows:

The two major hard-tech industries of new energy vehicles and humanoid robots possess a high degree of structural similarity. The developmental milestone for humanoid robots in 2026 is analogous to that of new energy vehicles during 2012-2014, with the latter's industrial evolution providing a clear phased framework for comparison. Both rely on mature large-scale manufacturing and leaps in AI algorithms. China's new energy vehicle sector, driven by national strategy, experienced explosive growth, evolving from policy-driven to market-driven and from technology-focused to ecosystem-focused. By 2026, robotics technology will have just crossed the "usable" threshold, with unprecedented policy support and capital enthusiasm, similar to the characteristics of the new energy vehicle sector after the launch of the Model S, although a closed-loop business model has not yet been formed.

While the humanoid robot and new energy vehicle industries have comparable developmental stages, their industrial essences differ. Intelligence is the core industrial anchor point for the former, comparable in importance to the power battery in the new energy sector. From 2008 to 2020, the core challenge for the new energy vehicle industry was overcoming the physical and chemical limits of power batteries. China leveraged economies of scale to achieve significant battery cost reductions, establishing a "battery is king" hardware investment logic. Currently, humanoid robots are analogous to the 2012 new energy vehicle sector; this hardware logic is only temporarily correct. The core challenge is the "intelligence deficit," and hardware bodies are rapidly commoditized as supply chain costs plummet. The industry's core lies in embodied intelligence, with service differentiation being the key value driver, making the embodied intelligent brain the core moat.

Hardware and intelligence are not opposing forces. By 2026, core hardware will still have significant room for iteration, and the two will form a virtuous cycle where "intelligence defines hardware, and hardware nourishes intelligence." The direction of hardware iteration will be dynamically defined by the demands of intelligence. Data is the core resource in the era of embodied intelligence, comparable to lithium ore. The capacity for collection and efficient production determines the upper limit of models, making the data industrial chain a key investment direction. Embodied intelligence faces a severe Physical AI data bottleneck. The trillions of physical interaction data points required for VLA models are vastly different from existing public datasets of millions, with the lack of proprioceptive data being the greatest challenge. Companies in the industry are competing for data mining rights through data collection centers, VR teleoperation, and motion capture, but acquiring such high-value data currently involves high costs and low efficiency.

Data factories are the core starting line for robotic intelligent agents. Enterprises capable of building large-scale teleoperation data collection pipelines at low cost and high efficiency will build a deep moat, forming a positive cycle of "data-capability-orders." Simulation technology and synthetic data have become important accelerators for data production, with initiatives like NVIDIA's Isaac Lab and synthetic data startups already taking the lead in this area. From an investment perspective, beyond basic hardware suppliers, potential beneficiaries worthy of attention include players in the data industrial chain, such as data collection service providers, simulation platform ecosystem partners, and scenario operators.

Core risks include fluctuations in raw material prices, geopolitical risks, and industry recovery falling short of expectations.