AI asset allocation is at the right time: Deshi-B (02526) value highlighted
The Zhitong Finance App learned that recently, the Hong Kong stock technology sector has continued to adjust, and the valuations and market expectations of AI-related companies have been re-examined. As far as long-term funding is concerned, this round of adjustments proposes a more important task: re-identifying assets that can continue to accumulate competitiveness in the development of the global AI industry and transform technical capabilities into commercial revenue.
AI should continue to be a key focus in long-term asset allocation. Among them, the model connects computing power investment with application requirements, and plays a key role in transforming computing resources into production capacity. As the industry enters the commercial realization stage, the configuration value of companies with autonomous models, professional data systems, and delivery capabilities is worth further scrutinizing.
DESHI-B (02526) is providing such an observation sample. To understand the value of this company, it is necessary to combine the medical imaging software and equipment business with the general medical imaging base behind it, and re-evaluate the growth space of a medical imaging model company.
Industrial logic iteration: from parameter competition to vertical scenario value implementation
The continued deepening of global AI competition provides an industrial context for this judgment. Currently, AI is simultaneously advancing along with infrastructure expansion, model capability improvement, and application penetration. Reasoning efficiency has improved and usage costs have been reduced, making it possible for more tasks to use AI on a large scale. For enterprises, the focus of investing in AI is also gradually becoming more specific: whether it can improve R&D efficiency, reduce delivery costs, and solve problems that are difficult to handle with the original technology.
This change makes the model layer a value link worth continuing attention. Excellent models can serve multiple customers and tasks, and enable early development to form the ability to be used over and over again. For model companies that master industry data, professional evaluations, and business processes, every delivery is an opportunity to increase the accumulation that can be reused for subsequent projects.
China's AI industry is taking a more prominent place in this process. Goldman Sachs's recent research summarizes global AI competition into two major ecosystems dominated by the US and China, and focuses on the development of China's industrial chain from infrastructure and models to applications. The competitiveness of the Chinese model is also being reflected in actual use.
According to Lyon research quoting data from the German OpenRouter platform, in September 2026, the Chinese model accounted for about 65% of the platform's token usage and 30% of the expenditure. This platform sample reflects the competitiveness of the Chinese model in terms of cost and usage requirements. The further transformation of usage into revenue and profit will be an important basis for judging the value of model assets in the next stage.
From a configuration perspective, China's AI opportunities are expanding along model capabilities to more professional scenarios. Language and video are two of the directions worth focusing on.
The language model mainly focuses on forming the ability to organize text, code, and knowledge, and serves for Q&A, programming, office work, and task execution. Image models require understanding spatial structure, organizational morphology, subtle characteristics, and continuous changes, and are directly used in medical and industrial inspection fields. The two types of competencies are being integrated at an accelerated pace, and the requirements for data integrity, recognition accuracy, and verification of results in professional scenarios continue to increase.
This trend is intersecting with the actual needs of China's healthcare system. From October 12 to 13, 2026, the National Health and Wellness Conference was held again after a lapse of 10 years, sending an important signal for the healthcare industry in the next 5 to 10 years. The conference emphasized strengthening operational guarantees for primary medical institutions in counties and districts, and proposed achieving safe operation of funds, healthy operation of hospitals, and reasonable burdens for the public.
This means that health-care reform requires improving the efficiency and accessibility of medical services while controlling costs and ensuring the sustainability of funds. Currently, high-quality doctors, patients, and technical resources are still concentrated in major hospitals, and grass-roots institutions have shortcomings in terms of talent and ability. In order for people to receive more reliable diagnosis and treatment services at their doorstep, in addition to equipment and financial 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 to connect the top three hospitals with primary care, and Deshi is a representative participant in this trend. If expert experience can be modeled and deployed at the grassroots level, primary doctors can obtain reviewable auxiliary tools, and patients also have the opportunity to receive services closer to the top three hospitals at their doorstep.
This is especially true of medical imaging. A CT scan includes a large number of spatially related slices. Pathological images need to preserve details at the cellular and tissue levels, and continuous imaging also involves time changes. The model needs to handle full inspection, preserve critical evidence, and allow professionals to review results. These requirements make medical imaging bases have independent technical and commercial value.
Pedestal + Platform Dual Wheel: Building a Core Barrier for Cross-Specialty Reuse
iMediimage, developed by Deshi itself, is revolving around this demand.
According to the company's 2026 interim report, iMedia uses a technical path of lower-level sensing adaptation, unified representation and inference to process different information such as two-dimensional images, CT and MRI three-dimensional data, full-slice pathology, cell morphology, medical videos and reports, and supports locating the output back to specific slices, image regions, lesions, or measurement results.
The commercial significance of this architecture is to promote cross-task reuse. Model development for different specialties can use the image understanding ability already accumulated by the base, and then combine professional data for training, evaluation, and deployment. Deshi's core model does not only deliver an image or a set of software, but rather provides medical institutions with the ability to use it sustainably based on large models, so that expert experiences can be replicated in more hospitals and grassroots scenarios after being digitized and modeled.
Therefore, when evaluating suitability, it should also consider the products it has delivered and how many professional tasks the same base can support in the future. The latter determines whether the company's growth space can break through the boundaries of a single product and a single disease.
Supporting this expansion is also a professional collaboration system. By the end of June 2026, the company had established cooperation with 99 hospitals and cooperated to carry out a total of 158 model projects, covering 61 disease directions. These figures reflect R&D cooperation and project accumulation, providing a basis for the development and verification of models in different professional tasks.
The company launched iMedloop in July, further connecting data processing, professional labeling, model training, evaluation, deployment and application feedback. At the time of disclosure in the interim report, the platform had gathered more than 3,000 professional participants and about 28.95 million labeling samples.
The value of medical imaging data depends on quality, interpretation standards, verification processes, and feedback mechanisms. Deshi's platform system is trying to organize these links so that the experience of forming a project can serve the next development. This not only helps improve model development efficiency, but also helps promote the professional experience of the top three hospitals to the grassroots level at a lower marginal cost. It is an important path for medical imaging model companies to build long-term competitiveness.
Gradual Verification of Commercialization: Opening a Window for Long-Term Revaluation
Commercial revenue has provided initial support for this logic.
In the first half of 2026, Deshi achieved revenue of about 109 million yuan, an increase of 21.0% year on year; model service revenue was about 94.54 million yuan, up 101.1% year on year, accounting for about 87% of total revenue; and overall gross margin was 74.1%. Model services have become a major driver of the company's revenue growth.
This business coverage model and technology licensing, MaaS cloud services, and localized delivery including hardware and implementation are adapted to customer needs in terms of data security, deployment environment, and usage methods, respectively. Multiple delivery forms enable the company to serve different types of institutions around the same technology base, and also provide conditions for subsequent expansion.
This business model is being further validated through regional platform cooperation. According to the company announcement, in September 2026, Deshi and Anji established a three-year cooperation framework. Anjianer is a provincial healthcare AI infrastructure platform in Zhejiang Province, and also undertakes the operation of a national artificial intelligence large-scale application pilot base (medical care). The two sides plan to build a provincial medical imaging AI cloud platform based on the Desi iMedia multi-modal medical imaging platform, explore the service model of “primary inspection, cloud interpretation, and provincial platform review”, and jointly develop medical data sets to explore medical data capitalization paths.
The announcement also revealed that the two sides will explore a model service payment model measured by tokens to lower the access threshold for medical institutions. This shows that tokens are not only a unit of measurement for generic models, but can also be a way for professional medical AI capabilities to be invoked and paid for. The annual number of outpatient visits in Zhejiang Province is about 759.6 million. According to conservative estimates in the announcement, the number of imaging tests may exceed 1 billion; if AI film reading services are implemented on a large scale in the future and a single service fee is estimated at 10 yuan, the corresponding potential market space could reach 10 billion yuan.
The direction of the company's investment is also clear. R&D expenditure in the first half of the year was about 64.12 million yuan, an increase of 67.4% over the previous year; as of the end of June, cash and cash equivalents were about 655 million yuan, and capital reserves after listing and financing provided resources for continuous R&D.
Looking at Deshi's room for revaluation from this, the key is how the market understands its asset attributes. Its cross-specialty reuse capabilities, professional development requirements, customer expansion efficiency, and continuous model service growth capacity based on a medical imaging base can be verified in more projects, and the company's revenue boundaries and cost structure may change further. The capabilities formed by foundation R&D can support more tasks; the experience brought by professional collaboration can improve subsequent delivery; and cloud and localized services provide different commercial implementation paths. Together, these factors form the core of Morality and Fit, which is worth reevaluating.
Judging from the long-term goals of the national health care system, the value of Deshi also lies in transforming “expert experience” into replicable model capabilities. If the model service can connect the top three hospitals, regional platforms, and primary institutions under the premise of compliance, there is an opportunity for high-level diagnosis and treatment experience to cover a wider range of people, help the grassroots improve imaging diagnosis capabilities, and improve the efficiency of the use of medical resources under the constraints of funds.
Therefore, the focus of the market's observation on Deshi should not stop at the scale of parameters or individual product sales, but should also focus on whether model capabilities can continue to enter the hospital workflow, whether professional data can continue to accumulate, whether regional platform cooperation can form actual projects, and whether model services can be converted into sustainable income. As these indicators are gradually realized, Deshi's value as an AI base for vertical medical imaging is expected to be more fully priced.