DeepSeek’s Rise: A Showcase of Chinese Innovation
The emergence of DeepSeek, with its distinctive architecture and economical technology, has garnered considerable attention from the global technology and venture capital communities. This year, DeepSeek’s ascent signifies an acceleration in the integration of artificial intelligence (AI) across a diverse range of industries.
During the “Technology Restructuring Value – 2025 Equity Investment Spring Forum,” hosted collaboratively by 21st Century Business Herald and the Tsinghua University School of Economics and Management Executive Education Center, thought leaders and experts from academic institutions, industrial sectors, and the investment arena convened to deliberate on the multifaceted implications of DeepSeek, AI’s application in robotics and healthcare, and the strategic approaches necessary for addressing the challenges precipitated by the proliferation of AI technologies.
‘China has demonstrated its ability to catch up in areas like new energy vehicles and large-scale models,’ observed Chu Xiaojie, General Manager of Strategic Cooperation at GalaxySpace. ‘DeepSeek’s emergence highlights the opportunities for differentiated development in China. A similar trend is also evident in the commercial space sector.’
Chu Xiaojie emphasized that China possesses the potential to rival the United States in traditional aerospace endeavors, building on decades of cumulative development. However, in the preceding 5–10 years, American commercial space enterprises, notably SpaceX, have instigated disruptive innovation within the industry, impacting both technological advancements and commercial models.
Pei Wanchen, Managing Partner at Xineng Venture Capital, posits that DeepSeek’s significance extends beyond its lower training costs compared to platforms like ChatGPT. Its true value lies in its capacity to democratize fundamental knowledge across various segments of Chinese society. Individuals can rapidly access comprehensive answers to their inquiries through DeepSeek’s intelligent search functionalities. However, Pei Wanchen cautioned that DeepSeek tends to offer more superficial rather than profound knowledge, as it primarily aggregates publicly available information, which inherently carries a temporal lag. Deeper understanding necessitates individual interpretation and subsequent judgment, processes that cannot be fully replicated by AI.
Notably, Xineng Venture Capital has invested in Muuxi Integrated Circuit, a key upstream player in the AI large model industry chain. Pei Wanchen elaborated that the team undertook extensive interviews and meticulous due diligence with nearly all domestic GPU chip companies before making the strategic decision to invest in Muuxi.
She stated her conviction that Muuxi’s early collaboration with Zhejiang University to tackle core technologies essential for domestically produced high-performance GPUs, and their efforts to construct a localized GPU supply chain coupled with a complete application ecosystem, provides the company with a distinct competitive advantage. When DeepSeek rapidly gained prominence, Muuxi’s technical team swiftly adapted and completed compatibility testing with Muuxi GPUs on the very day of the open-source release. This feat positioned Muuxi GPUs as the first domestically produced chips to achieve full compatibility with DeepSeek. In collaboration with Lenovo, Muuxi launched the inaugural domestic DeepSeek all-in-one solution, built upon the ‘Lenovo server/workstation + Muuxi training and inference integrated domestic GPU + independent algorithm’ architecture. This marks a crucial milestone for Muuxi within the domestic AI ecosystem and underscores local companies’ unwavering commitment to continuous innovation in large models and AI technologies. Muuxi is actively injecting renewed momentum into the intelligent transformation of diverse industries through its innovative breakthroughs in domestic AI infrastructure.
Zhang Miao, Co-founder and COO of Lingbao CASBOT, underscored the critical distinctions between embodied large models utilized in robotics and more generalized models such as DeepSeek, ChatGPT, and Doubao. Simply deploying a generic large model on a robot would likely result in what he termed a ‘human-shaped speaker’ rather than a fully functional robot capable of performing complex tasks.
Embodied large models are typically trained using a carefully curated combination of virtual and real-world data. Models trained predominantly in simulated environments often exhibit reasonable performance when deployed on real-world robots, but significant gaps still exist between simulated and actual performance. To mitigate these discrepancies, the team supplements the training data with real-world operational data collected directly at the robot’s designated task site. This hybrid training approach significantly enhances the usability of embodied large models and the overall performance of the robot in real-world scenarios.
Zhang Miao noted a discernible trend this year toward the maturation and deployment of a diverse array of robotic products and enabling technologies. From a commercial perspective, Lingbao’s robots find application in various B-end use cases, including industrial manufacturing, emergency rescue operations, and underground operations, with plans for small-scale mass production slated for this year. In the C-end market, owing to ongoing policy and regulatory developments, the company primarily engages with the public through carefully managed B2C interactions.
Accelerating AI Applications in Healthcare
As the DeepSeek era continues to advance in the technology sector, AI applications in healthcare are experiencing a parallel acceleration, driving transformative changes in medical practices and patient care. Li Feiyu, Executive Director and Deputy General Manager of Yimai Sunshine, emphasized that the core challenge in medical imaging lies in effectively managing both the acquisition and generation of imaging data, as well as the subsequent analysis and interpretation of this complex data. DeepSeek has the potential to substantially enhance the efficiency of imaging data analysis and diagnostic effectiveness by leveraging deep learning techniques on multimodal imaging big data.
For instance, a chest CT scan performed for lung nodule screening typically generates over 100 individual images, making manual review by radiologists a labor-intensive and time-consuming process. Furthermore, minute lung nodules, often measuring only a few millimeters in size, can be easily missed during manual examination. AI-powered imaging tools, such as those leveraging DeepSeek’s capabilities, can rapidly and thoroughly analyze these images, identifying subtle lesions that might be overlooked by the human eye.
DeepSeek has also facilitated the adoption of innovative business models that combine imaging equipment with AI-driven services, enabling hardware manufacturers to generate ongoing revenue streams by selling AI services in addition to the initial product sale. Moreover, AI tools like DeepSeek can assist in generating structured reports, which effectively standardizes and normalizes imaging data, thereby improving its overall quality and consistency. This standardization also fosters the emergence of medical data trading and licensing services, creating new opportunities for data-driven innovation in healthcare.
Zheng Hongzhe, Vice President and President of Strategy and Marketing at Yuyue Medical, highlighted that AI technology, as exemplified by DeepSeek, demonstrates a robust capability to process multimodal data, making it particularly well-suited for applications in the home medical device sector.
Yuyue Medical primarily focuses on addressing three major chronic diseases: respiratory diseases, hypertension, and diabetes. These conditions frequently overlap, resulting in the generation of comprehensive and complex data sets. Yuyue is continuously exploring and implementing new technologies to augment conventional single-point data with additional dimensions, thereby enhancing the accuracy and reliability of the data collected.
Secondly, the vast volume of data generated by home medical devices can be challenging for individual patients to interpret effectively, creating a pressing need to simplify complex data and make it more accessible and user-friendly. AI holds immense potential to address this critical challenge by providing personalized data insights and actionable recommendations. From this perspective, AI can play a significant supporting role, or even partially replace, roles traditionally held by professionals such as health managers, nutritionists, and exercise rehabilitation specialists in specific fields. For example, in a home setting, AI can provide non-diagnostic recommendations on comprehensive health management, nutrition management, and exercise health, empowering individuals to proactively optimize their lifestyles and improve their overall well-being.
The application of AI technology also introduces certain challenges that must be carefully addressed, such as the potential for hallucinations (the generation of factually incorrect or nonsensical information) and data security issues. Li Feiyu emphasized that data security is an unavoidable concern in AI applications, particularly given the sensitive nature of medical data. Training AI tools like DeepSeek requires large volumes of patient data, including personal information, clinical details, and imaging data. Any data breach or unauthorized access could severely harm patients’ rights and privacy.
‘We ensure the security and confidentiality of imaging data during storage, transmission, and use. We strictly adhere to national regulations by using data anonymization and various technical measures to ensure that data can be used legally, compliantly, and securely,’ stated Li Feiyu.
Wei Qiang, Professor and Head of the Department of Management Science and Engineering at Tsinghua University School of Economics and Management, who served as the host for the roundtable discussion, concluded that the forum’s central theme was ‘Technology Restructuring Value.’ However, regardless of how technology restructures value, it is imperative that it aligns with fundamental principles of human knowledge, common sense, morality, and ethical considerations. As technology is explored and implemented in diverse scenarios, human knowledge must be embedded to judiciously determine which tasks should be performed by humans and which can be effectively automated by machines. ‘A truly useful system for humanity is not a complete replacement of humans, but rather a system in which humans engage and guide at a higher level, which is where our true value lies.’ This collaborative approach, where humans and AI work in synergy, is essential for realizing the full potential of AI while safeguarding human values and ensuring responsible innovation.