Since September, over a dozen flagship global large language models, including GPT-6 Astra and Claude Fable 5.1, have been launched in quick succession, pushing the AI sector back into the market's trading spotlight. Regarding the allocation logic for Hong Kong-listed AI application stocks, Huatai Securities offered a clear perspective on September 4: until a new demand narrative is fully confirmed, the chip structure of the AI application track is significantly better than that of hardware, giving it a short-term relative advantage; moreover, the beta attributes of Hong Kong software stocks are stronger than those of A-shares, and the core anchor for stock selection remains the sequential acceleration of revenue growth, with earnings taking precedence.
Reviewing the evolution of this AI rally, the market is shifting from betting on model parameters to focusing on application deployment, leading to a clear divergence among Hong Kong AI application plays. Although the general-purpose large model track offers enormous imagination, it suffers from numerous participants, ongoing price wars, intensifying homogeneity, and long, uncertain earnings realization cycles; in contrast, the vertical AI track in highly regulated, high-barrier industries is becoming the more cost-effective direction in the AI application rally, thanks to clear commercialization paths, stable competitive landscapes, and confirmed earnings delivery rhythms.
Within the vertical AI track, medical imaging AI stands out as a high-quality segment that balances short-term earnings with long-term potential. Unlike consumer-grade AI applications, medical AI faces three core barriers—data compliance, clinical validation, and medical device regulation—which new entrants cannot quickly overcome with computing power alone, making the competitive landscape far superior to the general-purpose track. At the same time, with the improvement of medical insurance payment support and the release of hospital-side demand, medical AI is moving from pilot demonstrations to large-scale deployment, entering a period of earnings realization.
In this context, DIAGENS-B (02526) is precisely the scarce target in Hong Kong's medical AI track that perfectly matches the stock selection criteria of "high growth + strong moat + clear catalysts."
Earnings Side: Precisely Hitting the Core Metric of "Sequential Revenue Growth Acceleration"
The core of stock selection lies in earnings, and the growth quality of DIAGENS is in the top tier of Hong Kong's AI application sector. Its 2026 interim report shows that the company achieved revenue of RMB 108.7 million in the first half, up 21.0% year-on-year; among this, model service revenue reached RMB 94.541 million, surging 101.1% year-on-year, raising its share of total revenue to 86.9%. Compared to the 51.4% share of technology licensing revenue for the full year of 2025, both the growth rate and revenue share of the company's model business have accelerated sequentially, with the revenue structure completely transitioning from "device sales" to "model services." More importantly, this revenue is not one-off project income but service-based revenue covering model training, system integration, workflow adaptation, and ongoing maintenance, offering stronger customer stickiness and revenue predictability. While most Hong Kong AI application targets remain in the "concept validation with minimal revenue" stage, DIAGENS has already validated the feasibility of its business model with nearly 90% of revenue share, placing its earnings delivery at a leading level within the sector and fully meeting the core stock selection framework of "sequential revenue growth acceleration." In its initiation report, Kaiyuan Securities forecasts that the company's operating revenue will reach RMB 490 million, RMB 930 million, and RMB 1.72 billion for 2026-2028, maintaining near-doubling high growth for three consecutive years, and assigns a "Buy" investment rating; based on current share price, the PS (price-to-sales) valuations for 2026-2028 projected revenue are 82.6x, 43.7x, and 23.6x, respectively. From a horizontal valuation comparison, Zhipu and MiniMax in the general-purpose large model track currently have dynamic PS ratios exceeding 100x; meanwhile, large model targets in vertical professional tracks, given their higher industry barriers and commercial certainty, should command a valuation premium over general-purpose models. If we reference the median level of 100x PS for peers, this corresponds to a reasonable market value of nearly HK$60 billion based on the company's 2026 projected revenue, leaving ample room compared to current valuations; if the expected revenue for 2027 is achieved, the market value is likely to exceed HK$100 billion.
Moat Side: Deep Defensive Moat in the Vertical Track, Avoiding General-Purpose Model Competition
Unlike the logic of competing on parameters and price in the general-purpose large model space, DIAGENS's core moat is built on medical professional depth and regulatory compliance—barriers that general-purpose large model companies cannot quickly overcome with computing power alone. The company's self-developed iMedImage ® medical imaging foundation model is a vertical foundation natively trained on full-modality medical data, including two-dimensional imaging, three-dimensional CT/MRI, and pathological slices, rather than a fine-tuned version of a general-purpose model, giving it a generational advantage in professional metrics such as lesion detail preservation and clinical evidence chains. Built on this, its iMedLoop ™ full-process platform connects data annotation, model training, evaluation and release, and clinical feedback across the entire chain, reducing the data requirement for single-disease model development to 1/200 of the traditional level and compressing the cycle to 2-3 months, enabling industrialized mass production of specialty models. At the more critical regulatory moat level, the company's AI AutoVision ® secured the world's first NMPA Class III medical device registration certificate based on large model technology. This certificate is not just admission approval for a single product; it validates the replicability of the "foundation model—specialty model—compliant device" pathway, paving the regulatory road for the commercialization of more models in the future and constituting a long-term competitive moat.
Catalyst Side: Dual Events Land, Opening a Window for Value Re-rating
The company has recently welcomed two major catalysts that jointly drive value recovery. First, its inclusion in the Stock Connect program took effect on September 7. After the Stock Connect opens, the influx of mainland incremental capital will not only improve trading activity but also drive a re-rating of the valuation system from a small-cap Hong Kong biotech stock to a vertical AI platform company, repairing the previous liquidity discount. Second, the official unveiling of a joint laboratory for general AI and medical applications, co-built with The Hong Kong Polytechnic University, marks a clear trend of the company positioning in next-generation technology as global large models upgrade toward agent-based systems: leveraging world-class frontier research capabilities combined with its own industrial deployment strengths, it aims to jointly explore new paradigms of agent participation across the entire medical R&D process, potentially upgrading the company's technical moat from "model mass production capability" to "R&D paradigm advantage," opening up long-term growth ceilings.
Returning to the stock selection logic for Hong Kong AI applications, as the market stands now, the stage of pure concept speculation has passed, and capital is increasingly concentrating on targets with earnings, moats, and catalysts. DIAGENS offers both the earnings validation of doubling model service revenue and a deep moat in the medical sector, plus the dual catalysts of Stock Connect inclusion and technology upgrades, making it a rare and certain allocation target in Hong Kong's current AI application track.
Comments