After Tempus's surge, the AI4S value of Deshi Technology (02526) is being rediscovered

Title context: After Tempus's surge, the AI4S value of Deshi Technology (02526) is being rediscovered Text: On September 17, the intersection of AI and life sciences once again became a focus of the capital markets. At the Hong Kong market close, Jitai Technology (07666) rose 12.70%, and INSILICO (03696) rose 6.86%; during U.S. trading, Tempus AI (TEM.US) was once quoted at $81.12, up 15.93%. On the same day, DIAGENS-B (02526) closed up 0.39%. A single day's gain or loss cannot summarize a company's value, but it does provide an observation window worth noting: as the market begins to pay higher prices for medical data, scientific workflows, and AI platforms, is the capability Deshi is building still undervalued? Tempus's operating data provides a clue. The company's Data and Applications business revenue reached $93.2 million in the second quarter of 2026, up 28% year over year; the company disclosed that its Insights-related business grew 36% year over year. New related licensing signings in the quarter were approximately $200 million, which represents signed contract value, not all of the revenue recognized in the period. On September 15 at Morgan Stanley's Global Healthcare Conference, Tempus management emphasized that the moat comes from hospital connections, compliance agreements, data pipelines, and the matching of clinical records with molecular, pathology, and imaging information. The company subsequently announced that over the next several years it will build a platform containing 100,000 whole genomes linked to longitudinal clinical information, with a later goal of expanding to 1 million. What the market is focused on is shifting from a single AI product to a system that continuously produces high-quality medical data and model capabilities. This is precisely the core direction of AI4S (AI for Science). In the industry's common definition, AI4S connects machine learning with scientific data, mechanistic knowledge, simulation computing, and experimental validation to improve the efficiency of hypothesis generation, prediction, and knowledge discovery. AlphaFold 3 uniformly predicts interactions among multiple types of biomolecules, and GNoME demonstrated a path from candidate discovery to stability assessment; China's "AI Plus" initiative also lists artificial intelligence and scientific research as a key direction. AI4S must ultimately be confirmed through reproducible validation and continuous iteration, which is why data quality, expert knowledge, and feedback loops become infrastructure. Studying Deshi along this line reveals a company closer to the essence of AI4S. Deshi organizes real medical problems, imaging data, expert judgment, model training, evaluation, and application feedback into the same production chain. iMedLoop launched on July 4, 2026, connecting data access, professional annotation, review and quality control, training and evaluation, deployment and release, and application feedback; iMedStudio supports AI pre-annotation, expert revision, multi-person result comparison, and dispute arbitration, and the relevant functions have been used in clinical research and internal R&D projects. This process precipitates the tacit knowledge in medical practice into data and model development capabilities that can be reviewed, trained, and reused. Models assist experts in processing materials, experts correct model errors, and within the scope of authorization, quality-controlled data can enter training and validation; in the face of false positives or borderline cases, the platform can design targeted samples accordingly and use independent testing in projects to evaluate the effect of improvements. Every real task has the opportunity to accumulate task definitions, data standards, quality control rules, and deployment experience. As of the evening of September 17, the iMedLoop official website page showed approximately 29.015 million annotated samples, 466.8 TB of images, 222 active tasks, and 3,172 certified experts. The user agreement clearly states that uploaded data does not involve a transfer of intellectual property rights, and use and benefits depend on specific authorization; at present, it still operates on gifted credits for trial use, and paid services will be announced separately. Deshi's core competitiveness lies in organizing, governing, and transforming data, providing a foundation for long-term collaboration with hospitals, experts, and R&D institutions. Deshi already has an observable commercial foundation. In the first half of 2026, the company's model service revenue was 94.541 million yuan, up 101.1% year over year, accounting for 86.9% of total revenue. iMedLoop launched in July, later than the first-half reporting period. After platformization, it is necessary to observe whether model iteration can bring more usage and payment, whether data and tools can reduce development and delivery costs, and whether the same platform can be reused across specialties. These three paths correspond to revenue, cost, and business expansion space, and also constitute the most imaginative part of Deshi. As tasks increase, the platform may accumulate methods for continuously improving models; as methods are reused, the time for new problems to become usable tools is expected to shorten, and the collaboration efficiency of the data and expert network is also expected to improve. Deshi is building the closed loop that AI4S values: "discover problemsorganize datatrain modelsvalidate effectsredeliver." Expert feedback, data growth, and retraining themselves cannot prove that autonomous recursive self-improvement has been achieved, nor can they replace independent testing and deployment monitoring. The evidence supporting a valuation re-rating will come from model iteration results, licensing cooperation, sustained payment, and cost efficiency. Tempus's market performance reminds the market that only when high-quality medical data enters real workflows and generates commercial returns does it have a stronger basis for obtaining higher pricing. What makes Deshi worth re-examining is precisely the next-generation model production system being built beyond its existing products. If one is looking in the AI4S track for a target that is close enough to scientific problems and also has platform-based imaginative space, then based on public information, Deshi possesses the key conditions to become the purest and most imaginative target. Its "purity" comes from AI acting directly on medical research tasks; its "imagination" comes from the fact that every real task may become the starting point for the next round of model capability and commercial efficiency.