Digital News

　　On July 28, the 2025 World Artificial Intelligence Conference (WAIC 2025) concluded in Shanghai. The global event, which gathered over 1,500 representatives from more than 70 countries and regions, highlighted the profound shift of artificial intelligence (AI) from technological breakthroughs to real-world industrial applications. From addressing the reliability challenges of large models in vertical industries to the deep integration of AI with real-world applications and the collaborative development of open-source ecosystems and standardized systems, AI is reshaping diverse industries with a more practical approach.

　　Large Models Moving Towards Vertical Industry Applications

　　The development of large models has moved beyond the broad phase of parameter competition to achieve precise breakthroughs in vertical industries. Overcoming reliability bottlenecks through engineered systems has become an industry-wide consensus. At this year’s WAIC, several companies showcased innovative approaches to turning technology into specialized productivity, highlighting the new paradigm of “intelligence + engineering” as a dual-driving force.

　　Ant Misuan announced an open-source advanced program framework for trusted applications of large models, specifically targeting reliability issues in specialized scenarios such as healthcare and finance. Wei Tao, Chairman of Ant Misuan, stated, “The key to ensuring the reliability of large models lies not in the models themselves, but in the integration of engineering and intelligence.” The framework is now implemented throughout the entire financial risk control process, automating workflows and transforming traditionally manual, time-consuming tasks into intelligent orchestrations, greatly enhancing both efficiency and objectivity.

　　Ant Digital Technologies has launched the Agentar-Fin-R1, a large financial reasoning model designed to address the specific challenges of the finance industry. Built on Qwen3, this model has outperformed its competitors in authoritative evaluations such as FinEval 1.0. Its main strength lies in the creation of a comprehensive financial task system, encompassing 6 major categories and 66 subcategories of scenarios. Leveraging a vast, hundred-billion-level specialized corpus and weighted training algorithms, it ensures precise reasoning for complex business processes. Zhao Wenbiao, CEO of Ant Digital Technologies, noted that “The in-depth application of large financial models will become a key factor in enhancing the competitiveness of financial institutions.”

　　The AI innovation of the social platform Soul App in the social space highlights its ecological diversity. Its proprietary end-to-end full-duplex voice call model bypasses the traditional Voice Activity Detector (VAD) system, allowing AI to autonomously control the pace of conversations. In AI-hosted group chat party scenarios, the model naturally energizes the atmosphere and encourages user engagement, greatly enhancing the interactive experience. Soul’s CTO, Tao Ming, said, “We are committed to empowering connections with AI, creating engaging scenarios, and enhancing user experiences to bring them a sense of happiness and belonging.”

　　The “Yinchao” AI music creation app, developed by the emerging large-model company Ziyou Liangji, has expanded the boundaries of the content ecosystem. The product is built on a self-developed music large model utilizing an AR + NAR architecture, enabling the generation of complete songs from text, images, or melodies. Jiang Tao, CTO of Ziyou Liangji, noted, “Yinchao enables every moment of inspiration to be transformed into professional music works.” At the exhibition, interactive experiences like “Selfie to Song” gave participants the chance to experience how AI is reshaping the boundaries of music creation, demonstrating the influence of AI ecosystems in the cultural sector.

　　Accelerated Integration of AI and Hardware

　　AI is no longer just a concept confined to laboratories; it has now been integrated into real-world applications such as services, healthcare, and manufacturing through hardware solutions. At this conference, the commercialization of AI has become increasingly evident, with pathways ranging from bipedal robots to smart devices.

　　KEENON Robotics’ “Robot Workforce” captured the attention at the exhibition with the global debut of its bipedal humanoid service robot, XMAN-F1. In the central exhibition area, the robot showcased its impressive skills by performing tasks like popcorn making and beverage preparation, with remarkable precision in movement and environmental adaptability. KEENON Robotics has built a comprehensive service network across dining, healthcare, and theater sectors through the collaboration of multi-form robots. Its products are now available in over 600 cities across more than 60 countries, demonstrating the commercial potential of embodied intelligence.

　　With the theme of “Aggregation · Smart Transition,” H3C, a subsidiary of Unisplendour Co., Ltd., showcased how the LinSeer intelligent computing solution empowers hardware terminals. Making its global debut, the H3C UniPoD S80000 super pod product supports high-density deployment with 64 GPUs per rack and can scale up to 1,024 GPUs, providing computational power for trillion-parameter models. The company’s “Digital Health Experience Pod” provides round-the-clock health management via AI-driven digital avatars. In transportation scenarios, the urban rail data analysis AI agent achieved the real-time optimization of passenger flow scheduling and early fault alert.

　　Moore Threads’ “AI Factory” concept delivers foundational infrastructure for embodied intelligence at the chip level. The full-functional GPU, built on the self-developed MUSA architecture, combines AI computation, graphics rendering, and other capabilities. By leveraging FP8 mixed-precision technology, it achieves a 20% to 30% performance improvement in the training of cutting-edge large models. Zhang Jianzhong, CEO of Moore Threads, emphasized that to address the efficiency challenges in large model training caused by the rapid growth of generative AI, Moore Threads will introduce system-level engineering innovations to develop next-generation AI training infrastructure, aiming to create a “super factory” for producing advanced models.

　　United Imaging Intelligence’s “Human-Machine Showdown” in medical AI was a highlight at the exhibition. The brain metastasis AI system, developed in partnership with the Sun Yat-sen University Cancer Center, automatically detects lesions and generates imaging reports. It is now implemented in over 400 hospitals nationwide. During a live demonstration at Zhongshan Hospital Affiliated to Fudan University, three doctors used the AI-powered Chest Scan Multi-Check System to quickly identify abnormalities such as lung nodules and coronary artery calcification with just one click, boosting overall efficiency by 25%.

　　Ecological Development Reaching a New Depth

　　The competition in the AI industry has shifted from a focus on individual technologies to a contest of ecosystems. The standardization of basic components, such as open-source frameworks, data governance, and computing power collaboration, is essential for the industry’s large-scale growth. At this year’s WAIC, collaboration across government, industry, universities, and research institutes has led to a series of groundbreaking advancements in areas such as corpora, computing power, and models.

　　The Intelligent Computing Cluster Industry Development and Future Trends Research Report, jointly released by Enflame and multiple parties, guides the development of computing power infrastructure. Its third-generation product, the Enflame S60, was launched in 2024 and has been successfully commercialized on a large scale, with a deployment of 70,000 GPUs supporting over 300 application scenarios and the construction of five major intelligent computing clusters. Enflame CEO Zhao Lidong emphasized, “Open-source and openness are crucial for advancing the intelligent computing industry.” Enflame has worked with various partners to seek breakthroughs in the domestic intelligent computing sector, with a strong belief that collaborative efforts will pave the way for comprehensive breakthroughs.

　　At the Corpus Innovation and Development Forum, Kupasai (KPS) unveiled four key innovations, establishing a comprehensive production and governance framework for high-quality data. The company launched China’s first unified public service portal for corpus operations, built on the “1+N” model, enabling corpus access via agents and blockchain management. The updated internal corpus toolchain platform 2.0 features 403 functional modules, covering the entire process from multimodal data collection to intelligent data cleaning, and has been successfully implemented in industries like healthcare and finance. Shan Dongming, Chairman of the company, noted that corpus data must undergo a threefold transformation of methodology, infrastructure, and ecosystem to keep pace with the evolution of models. This approach has led the company to partner with universities such as Shanghai Jiao Tong University to establish joint master’s and Ph.D. programs in engineering, forming a “triple helix” of industry, universities, and research institutes, and pioneering a new model for industry-education integration in the corpus sector.

　　With the theme, “Harnessing Chip Power to Unlock New Possibilities, Shaping a Shared Future of AI,” MetaX comprehensively showcased its technological breakthroughs, industry contributions, and ecosystem collaboration in the AI computing power sector through both an exhibition booth and a themed forum. The company introduced its newly launched Xiyun C600 general-purpose GPU, which has established a closed loop of domestic industry and supply chains, covering everything from chip design and manufacturing to packaging and testing, while ensuring full control over its core technologies. Furthermore, MetaX, in collaboration with the China Electronics Standardization Institute, unveiled a joint laboratory aimed at advancing AI chip standards and promoting the development of an open-source ecosystem.

　　From technological advancements to the synergy of ecosystems, the new changes showcased at WAIC 2025 highlight the industry’s shift from being “technology-driven” to “value-driven.” As large models find reliable implementation paths in vertical industries, as embodied intelligence creates real-world service value, and as open-source standards build collaborative innovation networks, AI will transition from the laboratory to the core of industry, becoming a new quality productive force driving economic and social transformation.