CITIC SEC: AI+Medicine Global Innovation Accelerates, Focus on Two Main Themes of Beneficiary Targets.

CITIC SEC released a research report stating that the medical vertical large models, with stronger understanding, generation, and multimodal capabilities, have expanded the AI + medical market space in more complex scenarios through the two paths of improving quality and increasing efficiency. Overseas enterprises have made rapid progress in commercialization, achieving stable income in the B-end market and actively laying out C-end products. Focus on two main lines of AI + medical benefits: 1) From the perspective of technological barriers, explore enterprises with model capabilities and scarce data competition barriers. 2) From the perspective of commercialization landing, explore enterprises with a solid B-end customer base, mature products in areas and scenes such as electronic medical records clinical decision-making, electronic medical records, and speech recognition. CITIC SEC's main points are as follows: Large models drive the advancement of AI diagnosis and treatment technology, empowering all scenarios inside and outside the hospital. According to official information from iFlytek Medical, Medical Union, Tencent Health, Baidu AI Open Platform, AI has been able to empower the medical diagnosis and treatment industry in all scenarios, including pre-diagnosis (AI + triage/guidance, AI + preliminary consultation), in-diagnosis (AI + electronic medical record quality control, AI + diagnostic assistance), post-diagnosis and off-site care (AI + rehabilitation management, AI + home health care). The global AI-assisted medical industry has gone through three development stages: technological accumulation, traditional AI models, and Transformer models. AI small models can empower medical diagnosis and treatment based on natural language processing and image recognition. There is still an application market in the field of assistive diagnosis of specialized diseases and medical image recognition. Based on Transformer architecture training and fine-tuning, medical large model products are significantly stronger than traditional AI models in understanding, generating, and multimodal capabilities, complementing them, thereby achieving comprehensive diagnosis and treatment assistance in more complex scenarios. With the continuous improvement of underlying model technology capabilities, in 2023 Google launched Med-PaLM 2, whose model base PaLM 2 has reached a parameter scale of 340 billion; Med-PaLM 2, GPT-4 Medprompt, and Med-Gemini scored 86.5, 90.2, and 91.1 respectively on the mainstream medical question and answer test MedQA, surpassing the average level of human experts. Overseas market: significant advantage in model underlying capabilities, medical Copilot is the mainstream application direction According to data from Global Market Insights, the global AI medical health market is expected to reach approximately $14.4 billion in 2023 and is expected to reach $281.2 billion by 2032, with a CAGR of approximately 39.1% from 2023 to 2032. According to NVIDIA's disclosure at the 2024 GTC conference, the company's healthcare business has generated over $1 billion in revenue in the 2024 fiscal year, 2-3 years ahead of its target. One of the strongest medical vertical large models in the world, Med-Gemini, can handle a wide range of medical data modal information such as texts, images, videos, and biological signals. According to Google's official website, in seven benchmark tests, Med-Gemini surpassed all SoTA models and has a 44.5% advantage over the highest level among the various GPT-4 variant models. Application-wise, although there is a clear advantage in model capabilities abroad, proactive application scenarios have not yet been developed compared to domestic ones. In the US market, AI + diagnosis and treatment in the B-end scenario, especially in the doctor Copilot scenario, the business model is relatively mature. AI-assisted diagnosis and treatment companies like Tempus provide various interconnected AI applications to medical staff, and in February 2025, they launched the C-end product Olivia Health in the US market. Tempus announced that it is expected to achieve positive EBITDA in 2025. Suppliers of AI medical documentation assistance products Nuance and AI remote health management service provider Sword Health also have stable income. Domestic market: achieving commercialization first in B and then in C, two major demands drive market expansion The number of AI medical large models released by domestic enterprises has exceeded 50, and all of them adopt the same first-in-B and then-in-C strategy as the US market at the application landing level. In the B-end scenario, the coverage includes AI clinical diagnosis and treatment decision support, electronic medical record content quality control, etc.; in the C-end scenario, AI chronic disease/health management services represented by WeDoctor Holdings and AI post-diagnosis management systems represented by XUNFEIHEALTH have successfully achieved commercialization landing through the B2B2C model. The application landing of large models in domestic AI-assisted diagnosis and treatment has two driving factors: 1) Scarcity of primary medical resources: According to data from the National Health Commission, the proportion of visits to primary medical institutions in 2013 accounted for 59.06% of the total visits, but by 2023, it had decreased to 51.73%. General practitioners account for only 13.1% of registered physicians at the primary level and have become mainstream among physicians aged 45 and older with specialist qualifications and below. According to an announcement by Iflytek Co., Ltd., the company's Anhui primary medical assistant product can increase the standardization rate of electronic medical records from 5% to 89%, and the reasonable diagnosis rate from 70% to 88%. 2) AI diagnosis and treatment intelligent products can improve the efficiency of diagnosis and treatment in a cost-effective way: According to calculations, large models can compress the average consultation time per person through technological means, potentially driving up the average annual medical resource value of tertiary hospitals by millions of RMB, and that of secondary hospitals by hundreds of thousands of RMB, making it highly cost-effective compared to project costs. Market space: high-growth blue ocean areas, GBC end-commercialization landing in order AI large model diagnosis and treatment intelligent applications belong to the incremental demand in medical IT. For example, in the tender for the informationization construction of Nenjiang People's Hospital in December 2024, the hospital plans to invest up to 18 million RMB in AI-related system construction, accounting for 45% of the total budget for a single informationization construction, reflecting the urgent need for AI capabilities at the hospital end. According to data from the National Health Commission, the National Bureau of Statistics, and the procurement network, it is estimated that from 2025 to 2029, the cumulative market space for G-end large model diagnosis and treatment intelligent projects will reach 4.7 billion RMB, with a CAGR of 18% from 2024 onwards; the cumulative market space for B-end will be 14.2 billion RMB, with a 41% CAGR during the same period; Meanwhile, AI chronic disease management/health management and post-diagnosis management in the C-end have successful commercialization landing through the B2B2C model.The theory market scale of the large market is expected to reach 72.2 billion yuan.Risk factors: risks of customer expansion falling short of expectations, risks of policy changes for the AI-assisted medical industry, risks of further intensifying market competition, risks of macroeconomic fluctuations, risks of industry and corporate reputation, risks of AI technological advancements falling short of expectations.