The concept of "AI for Science" has rapidly evolved from an academic buzzword into a major structural theme capturing the attention of capital markets since 2025. As the capabilities of large models extend beyond general conversation into specialized fields like molecular dynamics, weather simulation, and evidence-based medicine, a critical question emerges: who will build the underlying AI infrastructure for scientific research? ABLE DIGITAL (02687) is providing an answer that is being repeatedly validated through practical projects with a group of China's top universities and research institutions.

This company, with its long-term focus on knowledge technology, has established deep collaborative relationships with leading institutions including Peking University, Zhejiang University, Nanjing University, Fudan University, University of Science and Technology of China, Tianjin University, and Harbin Engineering University. Instead of offering generic tools, ABLE DIGITAL has chosen a path of "delving into specific disciplines"—it does not simply package a large model as a research assistant. Instead, it starts from the knowledge graphs, data structures, and reasoning logic of particular scientific fields, working alongside frontline scientists to build dedicated AI-powered research engines.

AI for Science in Practice: A Snapshot Across Five Disciplines

At the Peking University Health Science Center, ABLE DIGITAL co-established the "Peking University Medical Future Learning Center" with the university. This is not a nominal project. The company deployed its FDE (Frontline Deployment) teams directly into clinical settings, working side-by-side with educators and medical experts to fine-tune models, validate data, and optimize workflows. Together, they built a "Source Graph" system covering multiple disciplines like human anatomy, pathophysiology, and immunogenetics, based on a self-developed medical discipline-specific large model and a Multi-Agent knowledge production architecture. The core feature of this system is "evidence-based"—every AI inference must be traceable, and generated conclusions must come with clear literature citations. In medicine, this is not a technical choice but a fundamental requirement.

At Zhejiang University, ABLE DIGITAL was deeply involved in constructing the knowledge graph for the AIM mechanics large model under the "101 Plan" for mechanics, and in developing the physical-AI simulation science experiment modules. This model covers 2,026 mechanics knowledge points and 4,189 knowledge relationships, integrates with experimental scenarios like hypersonic wind tunnel physical-AI simulations, and is equipped with a proprietary solver. Researchers can input natural language queries, and the system outputs internal force distributions and stress cloud diagrams. Abstract information like mechanical simulation data and robotic arm motion parameters is transformed into interactive visualizations, representing a systematic reconstruction of the underlying knowledge production method.

At Nanjing University, ABLE DIGITAL is deeply participating as an industrial infrastructure provider for AI for Science in the atmospheric science "101 Plan" led by Academician Tan Zhemin of the Chinese Academy of Sciences. This project involves 28 universities and research institutes. At the mid-term progress meeting in January 2026, "AI-Enabled Atmospheric Science" was listed as one of two core discussion topics. ABLE DIGITAL's physical-AI platform—which integrates text, speech, vision, and 3D generation technologies and has been connected to the World Labs world model—is supporting the construction of a new research ecosystem for disciplines like dynamic meteorology.

At Fudan University's Department of Nuclear Science and Technology, ABLE DIGITAL collaborated with nuclear science experts to build the "Nuclear + X" Smart Science and Education Platform. An AI large model systematically analyzed 31,147 key research papers, identifying 15 key industrial research areas, 102 research directions, and 587 research hotspots. It also performed an intelligent comparison of curricula and research systems from 13 domestic and international peer universities, generating 65 suggestions for discipline optimization. This is not a static research report but a dynamic AI algorithm system for monitoring research trends and industrial needs.

In frontier basic science, the collaboration between ABLE DIGITAL and the University of Science and Technology of China is also noteworthy. The company has transformed the study of Transient Luminous Events (TLEs) in the middle and upper atmosphere—including international frontier topics like red sprites, blue jets, and gigantic jets—into interactive experiments based on physical-AI simulation. Complex phenomena difficult to observe and model precisely with traditional methods, such as modeling discharge processes driven by atmospheric electricity laws, calibrating optical characteristic parameters, and analyzing the impact of atmospheric density at different altitudes on luminescent color, are recreated as scenarios researchers can repeatedly explore on a digital platform.

Across a broader disciplinary landscape, ABLE DIGITAL has partnered with Tianjin University (the "燧明" large model for optics), Harbin Engineering University (the "兴海" large model for naval architecture and ocean engineering), Beijing Forestry University (the "智源" large model for soil and water conservation), and Chongqing Medical University (the "干源" large model for medicine), forming a matrix of discipline-specific large models. Multiple researchers interviewed expressed a similar view: the value of these models lies not in competing with general large models on parameter scale, but in their deep adaptation to the specific data structures and research paradigms of individual disciplines.

Three-Layered Competitive Moat: Data, Systems, and Knowledge

Market analysts suggest that understanding ABLE DIGITAL's competitive moat requires moving beyond traditional SaaS valuation frameworks. In an era of exponential AI iteration, the company is simultaneously building barriers on three fronts.

The first layer is the data barrier. The company's vast repository of multimodal data is not publicly scrapable internet text. It consists of equipment operational fluctuation curves from physical experiments, layer-by-layer annotated anatomical structures from virtual dissections, computational fluid dynamics parameter distributions from hypersonic wind tunnel simulations, and scientifically calibrated spectral features from atmospheric discharge phenomena. Acquiring and labeling this data requires repeated verification by domain experts and AI systems. "The time and difficulty of this process far exceed that of pure text data engineering," noted several interviewees, who consistently used the same term to describe the value of this data: "irreplaceable."

The second line is the completeness of the technical system. Supporting the above scenario-based implementations is the company's continuously developed full-stack AI capabilities. In 2026, ABLE DIGITAL formally integrated the World Labs world model (Marble WM), adding to its existing capabilities in text large models (LLM), vision-language multimodal models (VLM), visual-only perception frameworks (VOM), automatic speech recognition (ASR), text-to-speech (TTS), and 3D generation models (3DM). This has formed a comprehensive LLM+VLM+VOM+ASR+TTS+3DM+WM technology matrix. This full-chain capability configuration—"from perception to cognition to generation"—is not common in China's AI for Science sector.

The third layer is the business model barrier. ABLE DIGITAL's FDE (Frontline Deployment) model—deploying teams directly into research settings to co-develop models, validate data, and optimize processes with scientists—is heavier and slower than standardized product delivery. However, the accumulated discipline-specific know-how and client trust it builds cannot be replicated through API calls or cloud deployment. When a company completes the full cycle from data collection to model training to scenario validation in a vertical discipline, what a latecomer must catch up with is not a product feature, but an entire methodology system refined over years of collaboration.

From Collaborative Discipline Building to Value Transformation

While the industry continues to debate large model parameter scales and general-purpose capability boundaries, ABLE DIGITAL has chosen a slower but more solid path—entering the anatomy lab at Peking University Health Science Center, the mechanics simulation platform at Zhejiang University's Zijingang Campus, the supercomputing center for atmospheric science at Nanjing University, and the literature analysis workshop at Fudan's Department of Nuclear Science and Technology. Each path leads to the knowledge infrastructure of a vertical discipline. What connects these infrastructures is an AI foundation capable of serving the core needs of China's scientific research system.

The implementation of AI for Science is, in essence, a race to build knowledge infrastructure. It is not about who has the largest general-purpose model, but about who can transform decades of disciplinary knowledge into computable, inferable, and reusable scientific corpora, and who can convert the tacit knowledge in experts' minds into platform-based intelligent capabilities.

The series of collaborations between ABLE DIGITAL and institutions like Peking University Health Science Center, Zhejiang University, Nanjing University, Fudan University, among others, demonstrates a potential solution path: using multimodal data as the foundation, structured knowledge graphs as the engine, and an on-site co-development model as the methodology, to embed AI capabilities into real-world research scenarios. From evidence-based medicine to mechanics simulation, from atmospheric physics to interdisciplinary nuclear science, these solid implementation cases are collectively forming the key nodes in China's AI for Science industrial landscape. As the digital foundation for scientific research grows increasingly robust, the story of artificial intelligence empowering scientific discovery is truly entering its main chapter.