China has quietly scaled AI-powered medical diagnosis to serve over 1 million patients in real clinical settings, with some applications already processing millions of cases across dozens of hospitals. The National AI Application Pilot Base, launched in 2025 and led by Zhongshan Hospital, released its first major achievements this week, unveiling nine flagship medical AI applications and five core technological breakthroughs that suggest a fundamentally different approach to deploying AI in healthcare: building the infrastructure first, then scaling rapidly. Unlike the startup-driven AI drug discovery boom or the incremental adoption of AI scribes in American hospitals, China's pilot base is taking a centralized, infrastructure-first approach. The base has constructed an integrated platform covering computing power, medical data, large language models (LLMs), and clinical validation systems, creating what officials describe as a "solid foundation for scaled AI application in clinical medicine". What Are the Nine Medical AI Applications Already in Use? The pilot base released nine applications spanning multiple clinical domains. The most mature deployments show striking scale: a hepatobiliary tumor AI agent is now deployed across 122 medical institutions serving over 1 million patients, while a chest imaging diagnostic tool has processed over 2.5 million cases in approximately 20 domestic hospitals. These aren't pilot projects or limited trials; they're functioning as operational systems in the Chinese healthcare system. The applications cover a broad range of clinical needs: - Cardiovascular AI: Zhongshan Hospital's "Guanxin" cardiovascular AI agent, designed to assist in heart disease diagnosis and treatment planning. - Urology AI: Renji Hospital's urology AI agent, which has already served over 500,000 patients in clinical practice. - Hepatobiliary Tumor Diagnosis: An AI agent deployed across 122 medical institutions for liver and bile duct cancer detection and analysis. - Chest Imaging Analysis: A one-scan multi-diagnosis agent that has processed over 2.5 million imaging cases across approximately 20 hospitals. - Brain-Computer Interfaces: Huashan Hospital's invasive brain-computer interface device, which became the world's first to receive China's Class III medical device certificate. - Medical Device R&D: AI systems supporting the research and development of new medical devices. - Drug Discovery: AI-powered drug discovery applications accelerating pharmaceutical research. The scale of these deployments is notable because it suggests that AI medical applications can move beyond experimental settings into routine clinical use relatively quickly when proper infrastructure exists. The chest imaging tool processing 2.5 million cases represents not a theoretical capability but actual diagnostic work being performed in hospitals. How Is China Building the Technical Foundation Differently? The pilot base's five core technological breakthroughs reveal a strategy focused on independence and standardization. Rather than relying on imported AI models or computing infrastructure, the base developed a domestic computing power platform that achieves "world-class performance" while running "domestic models on domestic chips". This addresses a critical vulnerability in global AI healthcare: dependency on foreign technology and cloud infrastructure. The base also created six vertical medical large models tailored to specific clinical scenarios, rather than attempting to use general-purpose AI models for medical tasks. This specialization approach mirrors how medical practice itself works; cardiologists don't use the same diagnostic framework as urologists. The base built a national demonstrative medical AI data infrastructure, establishing standardized datasets that multiple hospitals can use to train and validate AI systems. Perhaps most relevant for global AI development, the base created MedBench 4.0, described as a "globally leading Chinese medical large model testing platform." This is essentially a standardized benchmark for evaluating medical AI performance, similar to how researchers use benchmarks like MMLU (Massive Multitask Language Understanding) to compare general AI models. Having a shared testing standard allows hospitals and researchers to compare different AI systems fairly. Steps to Understand How Medical AI Infrastructure Scales in Practice The pilot base's approach offers a roadmap for how medical AI moves from research to routine clinical use. Understanding these steps helps explain why some countries may see faster AI adoption in healthcare than others: - Build Shared Computing Infrastructure: Rather than each hospital purchasing its own AI systems, the pilot base created centralized computing power that multiple institutions can access, reducing costs and ensuring consistent performance across the network. - Standardize Medical Data: The base established a national medical AI data infrastructure, meaning hospitals contribute anonymized patient data to a shared pool that trains AI models benefiting the entire system, rather than each hospital training models in isolation. - Create Vertical Specialization: Instead of one general AI model, the base developed six specialized large models for different clinical domains, recognizing that cardiovascular diagnosis requires different AI training than urology or oncology. - Establish Testing Standards: MedBench 4.0 provides a common benchmark for evaluating AI performance, allowing hospitals to compare different systems and ensuring quality standards across deployments. - Integrate Clinical Validation: The platform includes built-in mechanisms for validating AI recommendations against actual patient outcomes, creating feedback loops that improve model accuracy over time. This infrastructure-first approach contrasts sharply with how AI adoption has unfolded in Western healthcare, where individual hospitals and health systems often implement AI tools independently, leading to fragmented data, inconsistent standards, and slower scaling. What Does This Mean for the Global AI Healthcare Race? The pilot base's achievements suggest that centralized healthcare systems may have structural advantages in scaling AI faster than decentralized systems. China's ability to coordinate across 122 hospitals using a single hepatobiliary tumor AI agent, or to process 2.5 million chest imaging cases through a standardized system, reflects the coordination advantages of a unified national healthcare approach. "Several applications have seen large-scale application: the hepatobiliary tumor AI agent is used in 122 medical institutions serving over 1 million people, while the chest imaging one-scan multi-diagnosis agent has processed over 2.5 million cases in some 20 domestic hospitals, greatly improving medical efficiency and precise medicine," stated Gu Jianying, chief of the pilot base at Zhongshan Hospital. Gu Jianying, Chief of the National AI Application Pilot Base at Zhongshan Hospital The pilot base also signed cooperation agreements with four Shanghai districts, eight research institutes, 28 hospitals, and 20 technology enterprises, creating what officials describe as a "government-industry-university-research collaborative ecosystem". This network approach suggests that scaling AI in healthcare isn't just a technical problem; it's an organizational and governance challenge that requires coordinating multiple stakeholders around shared standards and shared infrastructure. Looking forward, the base plans to advance achievement transformation, build a global medical AI innovation hub, and contribute to China's "Healthy China" initiative. The emphasis on "transformation" suggests that the pilot phase is ending and the focus is shifting to making these AI applications routine parts of clinical practice rather than experimental projects. For hospitals and health systems globally, the pilot base's model raises a practical question: can decentralized healthcare systems achieve similar scaling without centralized coordination? The answer may determine whether AI adoption in healthcare accelerates or remains fragmented by institution and region.
