Shanghai, China--(Newsfile Corp. - November 19, 2026) - Frost & Sullivan released the 2026 Blue Book on the Development and Applications of China's On-device AI Industry during the World Internet Conference Wuzhen Summit. Guided by the Artificial Intelligence Working Committee of the Internet Society of China, the Blue Book was jointly prepared by Frost & Sullivan, LeadLeo, JunHe Law Office and Jia Yuan Law Office.
Drawing on industry research, company interviews, expert insights and case studies, the Blue Book systematically introduces the technological development process and core characteristics of edge AI, focusing on the perception input layer, edge computing layer, edge model layer, and intelligent application layer. The Blue Book analyzes the main driving factors for the development of the on-device AI industry from the perspectives of technological progress, terminal demand, and industrial chain collaboration. The Blue Book also introduces the market development of on-device AI and the innovative application practices of representative enterprises and further explores the future development trends of on-device AI in terms of comprehensive task performance, agent collaboration, cross-device services and industry-specific solutions.
XFEON is selected as a benchmark case in the Blue Book, reflecting its technological capabilities and application progress in on-device AI computing, embodied intelligence and satellite computing.
Development Background of China's On-device AI Industry
Evolution from Cloud-Based AI to On-device Intelligence
As AI models continue to advance, AI computing is evolving from predominantly cloud-based processing toward a distributed architecture spanning cloud, edge and device environments. AI capabilities are increasingly being deployed across smartphones, personal computers, intelligent vehicles, robots, wearable devices and smart home appliances.
On-device AI enables part or all of data processing and model inference to be performed locally, helping reduce response latency and reliance on cloud computing resources while enhancing data privacy protection and enabling personalized services. It also allows intelligent devices to maintain selected AI functions under limited network connectivity, supporting real-time perception, decision-making and task execution.
Research Framework and Key Industry Insights
The Blue Book examines the on-device AI industry across five major areas: market definition and development, industrial value chain and value distribution, key terminal markets and application scenarios, competitive landscape and industrial ecosystem, and benchmark cases and development trends.
The research establishes a four-layer technological framework comprising the perception input layer, edge computing layer, on-device AI model layer and intelligent application layer. It further analyzes seven major application domains, including AI smartphones, AI PCs, robotics, industrial intelligence, smart wearables, intelligent vehicles and smart home appliances.
The findings highlight a shift in industry competition from standalone computing performance toward the integration of chips, model optimization, software tools and application capabilities. As AI applications extend into physical environments, computing efficiency, real-time responsiveness, system reliability and cross-platform compatibility are becoming increasingly important factors in commercial deployment.
XFEON Selected as a Benchmark Case in the On-device AI Industry
XFEON has developed integrated capabilities spanning on-device AI chips, computing platforms, model optimization and data processing. The Blue Book presents two representative case studies, demonstrating how these integrated capabilities support the deployment and continuous optimization of AI systems in real-world scenarios.
Case Study 1: Supporting Embodied Intelligence through On-device Computing and Data Capabilities
As embodied intelligence advances, robots are increasingly required to perform complex tasks involving multimodal perception, local inference and real-time decision-making. These applications place higher requirements on computing efficiency, memory capacity, response latency and system reliability.
XFEON has collaborated with leading embodied AI companies to establish the Ma'anshan Embodied Intelligence Innovation Center. Through its StarCore R Series computing products and XGAIA embodied intelligence data processing platform, the company supports robotic model development, data processing and iterative optimization.
The StarCore R Series delivers AI computing performance of up to 420 TOPS (INT8 equivalent) and supports up to 128 GB of memory. According to XFEON, under customer-specified testing conditions, its solutions reduced key task response latency by approximately 30% and improved the accuracy of typical task understanding and reasoning by approximately 50%.
The integration of on-device computing and data processing capabilities supports continuous model refinement and helps robotics developers improve the efficiency of AI deployment and application validation.
Case Study 2: Enabling On-orbit Intelligent Computing through the StarCore K Series
Satellite computing represents an emerging application of on-device AI, where computing systems must operate under stringent constraints relating to power consumption, available space, energy efficiency and environmental reliability.
As an ecosystem partner of the XingSuan Program, XFEON provided a domestically developed and independently controllable satellite computing solution based on its StarCore K Series for the Group 02 space-computing satellites of Guoxing Aerospace.
Through AI model compression and heterogeneous computing acceleration, the solution delivers computing performance of up to 300 TOPS and energy efficiency of up to 2.5 TOPS/W, improving the utilization of limited onboard computing resources.
The eight satellites in the Group 02 mission were launched from the Jiuquan Satellite Launch Center on October 14, 2026, and have since entered the in-orbit application validation stage. The project demonstrates the potential of domestically developed on-device AI computing technologies in space-based intelligent processing.
Industry Insights: Opportunities for On-device AI Chips in the Era of Physical AI
During the closed-door workshop on FlagOS-based heterogeneous computing and the industrial deployment of on-device AI, held alongside the World Internet Conference Wuzhen Summit, Gao Junxiao, Partner and Vice President of XFEON, delivered a keynote presentation entitled Intelligent Everything, Chip-Driven Future: Opportunities for On-device AI Chips in the Era of Physical AI.
Gao highlighted the evolution of AI from perception AI and generative AI toward agentic AI and physical AI. As AI systems increasingly move beyond information processing to perceive, plan and act in the physical world, intelligent devices are required to deliver more efficient, reliable and responsive computing capabilities.
This transition is reshaping on-device AI chip development. Model compression, low-bit quantization, heterogeneous computing, inference engine optimization and multimodal integration are becoming increasingly important in improving overall system efficiency. Rather than focusing exclusively on peak computing performance, chip manufacturers need to coordinate hardware architecture, model characteristics, software tools and application requirements.
According to XFEON, its mixed-precision quantization technologies can reduce model size by up to 85%, while limiting the loss in model accuracy to approximately 1%-2%. Combined with proprietary IP and hardware-level mixed-precision support, these optimizations can improve effective computing performance by approximately 1.5 times while keeping power consumption below 25 W.
The development of physical AI is expected to further increase demand for on-device computing technologies capable of supporting continuous perception, autonomous decision-making and real-time interaction across robotics, industrial systems and other intelligent devices.
Download Link: https://www.frostchina.com/content/insight/detail/6aeb056700e951eb8788c015?whitepaper=true
Contact Information
Contact: Esun Xu
Company: Frost & Sullivan
Website: https://www.frostchina.com
Email: esun.xu@frostchina.com
To view the source version of this press release, please visit https://www.newsfilecorp.com/release/324846
Source: Hmedium