Sinolink: Mid-term high-barrier and high-value segments of physical AI; bullish on data-collection gloves and dexterous hands

Sinolink issued a research report stating that under the integration of VLA (vision-language-action models) and world models, Siasun Robot&Automation is expected to be driven from single-step operations toward multi-step planning, failure recognition, and autonomous recovery. Data-collection gloves and dexterous hands are not substitutes for each other, but rather two complementary data-collection approachesthe "human end" and the "machine end." Sinolink is bullish that tactile demand from Siasun Robot&Automation will drive volume growth in both data-collection gloves and dexterous hands as data-collection equipment. Driven by the development of physical AI, data-collection gloves and dexterous hands are expected to become core incremental segments in the industry trend. Sinolink's main points are as follows: GPT-6 Astra fills the "brain" shortcoming of Siasun Robot&Automation; the next stop will be the evolution of System 1 driven by physical interaction data GPT-6 Astra succeeded in 49 of 50 episodes, a success rate of 98%, while the single 0.5 model achieved only a 36% success rate, proving that a general large model can serve as the System 2 (slow-thinking brain) of Siasun Robot&Automation. The long-standing industry constraint of "the brain isn't good enough" has reached an inflection point. Physical AI, represented by world models, extends the AI capability chain from "perceptionunderstanding" to "perceptionpredictionplanningcontrol." Under the integration of VLA (vision-language-action models) and world models, Siasun Robot&Automation is expected to be driven from single-step operations toward multi-step planning, failure recognition, and autonomous recovery. Industry consensus is gradually converging toward a hybrid architecture of "general reasoning + specialized operation"large models handle System 2 top-level cognition, while embodied models handle System 1 physical execution. For System 1 to evolve, it must be fed physical interaction data that simulation cannot provide. Data-collection gloves and dexterous hands are expected to become core incremental segments of physical AI What physical AI core needs is physical state datacontact force, tangential force, deformation, contact timing, and other data. These tactile data require real interaction collection and annotation. As of early 2026, the global stock of high-quality real physical interaction data was approximately 500,000 hours, while training a general embodied model requires at least 10 million hoursthe existing stock is less than 5% of demand, implying substantial medium- to long-term demand for tactile collection-related equipment. Core tactile data collection equipment includes data-collection gloves (human-end collection) and dexterous-hand teleoperation (machine-end collection). Among them, data-collection gloves are worn by humans and collect data directly, with lower cost than dexterous hands, and can enter various real scenarios for manual data collection, mainly for pre-training and tactile representation learning; dexterous hands are the source of real-machine tactile data for Siasun Robot&Automation, with core advantages being precise action labels based on the Siasun Robot&Automation body and no need for retargeting, leaning toward policy fine-tuning and real-machine validation. Data-collection gloves and dexterous hands are not substitutes for each other, but rather two complementary data-collection approachesthe "human end" and the "machine end." Sinolink is bullish that tactile demand from Siasun Robot&Automation will drive volume growth in both data-collection gloves and dexterous hands as data-collection equipment. Investment recommendations and valuation Driven by the development of physical AI, data-collection gloves and dexterous hands are expected to become core incremental segments in the industry trend. Related names include Ningbo Cixing, Dongguan Tarry Electronics, Zhejiang Fulai New Material, and Shenzhen Everwin Precision Technology. (1) Ningbo Cixing: a leader in computerized flat knitting machines, extending into flexible sensing fabrics based on 3D knitting technology. The company acquired Switzerland's Steiger in 2010 and signed an agreement in June 2026 to acquire assets related to the German STOLL brand, supplementing high-end knitting equipment, 3D braiding, and fabric design capabilities, and expanding into application areas such as smart wearables and flexible sensing. (2) Dongguan Tarry Electronics: a supplier of consumer electronics structural and functional components, with precision die-cutting, flexible material processing, and multi-material composite manufacturing capabilities. Its related technologies share commonalities with the flexible substrates, sensor fixation, and structural protective materials required for tactile data-collection gloves. (3) Zhejiang Fulai New Material: leveraging polymer materials and precision coating technology to develop flexible tactile sensors, with products already in volume delivery. In February 2026, it reached a procurement cooperation agreement with Lingxin Qiaoshou for 100,000 sets of tactile sensors; as of the end of August, cumulative deliveries exceeded 30,000 sets, mainly supporting industrial-grade dexterous hands, and expanding toward tactile data collection terminals. (4) Shenzhen Everwin Precision Technology: leveraging precision machining and multi-material molding capabilities, its products cover humanoid Siasun Robot&Automation dexterous hand transmission components, joint core transmission components, and torso skeletal structural components. It is also expanding into force-controlled joint modules, with related products having completed multiple rounds of sampling and testing, involving torque sensing, force feedback, and actuation control. Risk warnings Risk of Siasun Robot&Automation industry development falling short of expectations; risk of large-model progress falling short of expectations; risk of R&D progress in data-collection gloves and dexterous hands falling short of expectations.