The Hong Kong technology sector has been undergoing a sustained correction recently, prompting a reassessment of valuations and market expectations for AI-related companies. For long-term capital, this round of adjustment raises a more important task: identifying assets that can continuously build competitiveness in the global AI industry's development and convert technological capabilities into commercial revenue.
AI should remain a key focus in long-term asset allocation. Among these, models connect computing power investment with application demand, playing a critical role in transforming computing resources into production capacity. As the industry enters the commercial monetization phase, companies with proprietary models, professional data systems, and delivery capabilities deserve further scrutiny regarding their allocation value. DIAGENS-B (02526) provides exactly such a case study.
Understanding this company's value requires combining its medical imaging software and equipment businesses with the general-purpose medical imaging foundation model behind them, to re-evaluate the growth potential of a medical imaging model company.
Industrial Logic Evolution: From Parameter Competition to Vertical Scenario Value Realization
The continued deepening of global AI competition provides the industrial backdrop for this judgment. Currently, AI is advancing simultaneously along infrastructure expansion, model capability improvement, and application penetration. Rising inference efficiency and falling usage costs are making more tasks suitable for scaled AI adoption. For enterprises, the focus of AI investment is also becoming more specific: whether it can improve R&D efficiency, reduce delivery costs, and solve problems that previous technologies could not address.
This shift makes the model layer a value segment worthy of sustained attention. Excellent models can serve multiple customers and tasks, allowing early R&D investment to form capabilities that can be repeatedly called upon. For model companies that possess industry data, professional evaluation, and business processes, each delivery has the opportunity to add reusable accumulation for subsequent projects.
China's AI industry is forming a more distinct position in this process. Recent Goldman Sachs research summarizes global AI competition as two major ecosystems led by the United States and China, and highlights China's industrial chain development from infrastructure to models to applications. The competitiveness of Chinese models is also being demonstrated through actual usage. According to data from the OpenRouter platform cited by CLSA research, in September 2026, Chinese models accounted for approximately 65% of the platform's token usage and 30% of expenditure. This platform sample reflects the competitiveness of Chinese models in terms of cost and usage demand. Further converting usage into revenue and profit will become an important basis for judging the value of model assets in the next phase.
From an allocation perspective, China's AI opportunities are extending from model capabilities to more specialized scenarios. Language and imaging are two directions worthy of focused observation. Language models primarily form capabilities around text, code, and knowledge organization, serving Q&A, programming, office work, and task execution. Imaging models need to understand spatial structures, organizational forms, subtle features, and continuous changes, with direct applications in healthcare, industrial inspection, and other fields. The two types of capabilities are accelerating their integration, and the requirements of specialized scenarios for data completeness, recognition accuracy, and result verification are also continuously rising.
This trend is converging with the real needs of China's healthcare system. From October 12 to 13, 2026, the National Health and Wellness Conference was held again after a 10-year gap, releasing important signals for the healthcare industry over the next 5 to 10 years. The conference emphasized strengthening the operational support of county-level and grassroots medical institutions, and proposed achieving safe fund operation, sound hospital functioning, and reasonable patient burdens. This means healthcare reform needs to improve service efficiency and accessibility while controlling costs and ensuring fund sustainability.
Currently, high-quality doctors, patients, and technological resources remain concentrated in large hospitals, while grassroots institutions have shortcomings in talent and capabilities. To enable people to receive more reliable diagnosis and treatment services at their doorsteps, beyond equipment and funding support, it is also necessary to transform the experience of high-level experts into replicable and verifiable digital capabilities. AI is expected to become an important tool connecting tertiary hospitals with grassroots healthcare, and DIAGENS is a representative participant in this trend. If expert experience can be modeled and deployed to the grassroots level, grassroots doctors can obtain reviewable auxiliary tools, and patients can also have the opportunity to receive services closer to tertiary hospital standards at their doorsteps.
This is especially true for medical imaging. A CT examination contains a large number of slices with spatial relationships, pathology images need to preserve details at the cellular and tissue levels, and continuous imaging also involves temporal changes. Models need to process complete examinations, preserve key evidence, and allow professionals to review results. These requirements give medical imaging foundation models independent technological and commercial value.
Foundation Model + Platform Dual Engine: Building Core Barriers for Cross-Specialty Reuse
DIAGENS's self-developed iMedImage is developed around this need. According to the company's 2026 interim report, iMedImage adopts a technical path of bottom-layer perception adaptation, upper-layer unified representation and inference, processing different types of information including 2D images, CT and MRI 3D volume data, whole-slide pathology, cell morphology, medical videos, and reports, and supports locating outputs back to specific slices, image regions, lesions, or measurement results. The commercial significance of this architecture lies in promoting cross-task reuse. Model development in different specialties can leverage the imaging understanding capabilities already accumulated by the foundation model, and then combine them with professional data for training, evaluation, and deployment.
DIAGENS's core model is not just delivering a single image or a set of software, but providing medical institutions with continuously callable capabilities based on large models, so that expert experience, after being digitized and modeled, can be replicated to more hospitals and grassroots scenarios. Therefore, evaluating DIAGENS should consider both the products it has already delivered and how many professional tasks the same foundation model can support in the future. The latter determines whether the company's growth potential can break through the boundaries of a single product or a single disease type.
Supporting this expansion is also a professional collaboration system. As of the end of June 2026, the company has established cooperation with 99 hospitals, with a cumulative 158 model projects covering 61 disease directions. These figures reflect R&D cooperation and project accumulation, providing a foundation for model development and validation across different professional tasks. The iMedLoop platform launched by the company in July further connects data processing, professional annotation, model training, evaluation, deployment, and application feedback. As of the interim report disclosure, the platform has gathered more than 3,000 professional participants and approximately 28.95 million annotated samples.
The value of medical imaging data depends on quality, interpretation standards, validation processes, and feedback mechanisms. DIAGENS's platform system is attempting to organize these elements so that experience formed in one project can serve the next development. This helps improve model R&D efficiency and also helps promote the professional experience of tertiary hospitals to the grassroots level at lower marginal costs, which is an important path for medical imaging model companies to build long-term competitiveness.
Commercialization Gradually Validated: Opening a Window for Long-Term Value Reassessment
Commercial revenue has already provided preliminary support for this logic. In the first half of 2026, DIAGENS achieved operating revenue of approximately 109 million yuan, up 21.0% year-on-year; model service revenue was approximately 94.54 million yuan, up 101.1% year-on-year, accounting for approximately 87% of total revenue; overall gross margin was 74.1%. Model services have become the primary driver of the company's revenue growth. This business covers model and technology licensing, MaaS cloud services, and localized delivery including hardware and implementation, respectively adapting to customers' needs in data security, deployment environments, and usage methods. Multiple delivery forms enable the company to serve different types of institutions around the same technology foundation, and also provide conditions for subsequent expansion.
This business model is being further validated through regional platform cooperation. According to company announcements, in September 2026, DIAGENS entered into a three-year cooperation framework with Anzhener. Anzhener is Zhejiang Province's provincial-level medical health AI infrastructure platform and also operates the National AI Scaled Application Pilot Base (Medical). The two parties plan to rely on DIAGENS's iMedImage multimodal medical imaging foundation model to build a provincial-level medical imaging AI cloud platform, exploring a service model of "grassroots examination, cloud interpretation, provincial platform review," and jointly developing medical datasets to explore paths for medical data assetization. The announcement also shows that both parties will explore a token-based model service payment model to lower the threshold for medical institutions to access services.
This indicates that tokens are not just a unit of measurement for general-purpose models, but can also become a way for professional medical AI capabilities to be invoked and paid for. Zhejiang Province has an annual outpatient volume of approximately 759.6 million visits; based on conservative estimates in the announcement, imaging examination volume could exceed 1 billion times; if AI imaging reading services achieve scaled implementation in the future and the per-service fee is estimated at 10 yuan, the corresponding potential market space could reach the hundred-billion-yuan level.
The company's investment direction is also clear. First-half R&D expenditure was approximately 64.12 million yuan, up 67.4% year-on-year; as of the end of June, cash and cash equivalents were approximately 655 million yuan, and the capital reserves after IPO financing provide resources for continued R&D.
Looking at DIAGENS's reassessment potential from this perspective, the key lies in how the market understands its asset attributes. Its cross-specialty reuse capabilities built on the medical imaging foundation model, professional development needs, customer expansion efficiency, and ability to sustain model service growth — once these can be validated in more projects, the company's revenue boundaries and cost structure may undergo further changes. Capabilities formed through foundation model R&D can support more tasks; experience brought by professional collaboration can improve subsequent delivery; and cloud and localized services provide different commercial implementation paths. Together, these factors constitute the core of why DIAGENS deserves reassessment.
From the perspective of the long-term goals of the national healthcare system, DIAGENS's value also lies in transforming "expert experience" into replicable model capabilities. If model services can connect tertiary hospitals, regional platforms, and grassroots institutions under compliance premises, there is an opportunity to extend high-level diagnostic experience to a broader population, help grassroots levels improve imaging diagnostic capabilities, and enhance the efficiency of medical resource utilization under fund constraints.
Therefore, the market's focus in observing DIAGENS should not be limited to parameter scale or single-product sales volume, but should also pay attention to whether model capabilities can continuously enter hospital workflows, whether professional data can continuously accumulate, whether regional platform cooperation can form actual projects, and whether model services can be converted into sustainable revenue. As these indicators are gradually fulfilled, DIAGENS's value as a vertical medical imaging AI foundation model is expected to receive more adequate pricing.
Comments