Orient Securities has released a research report indicating that AI for Science (AI4S) is already making significant inroads across life sciences, Earth sciences, and energy materials, yet its overall industry penetration remains relatively low. The firm anticipates a transition into a new phase characterized by large-scale value realization and commercialization.

AI4S is actively empowering multiple disciplines and industries, including biopharmaceuticals, materials and energy, industrial simulation, meteorological and environmental science, and agricultural breeding. The rapid development and iterative progress of AI4S are creating notable investment opportunities across three key areas: scientific model innovation, the construction and operation of computational infrastructure, and downstream industrial applications. Here are the primary viewpoints from the brokerage.

AI4S Emerges as the 'Next Frontier' for Scientific Research and Large Model Evolution

AI4S represents a new paradigm that uses artificial intelligence to assist or independently complete scientific research tasks. It can comprehensively reshape the research workflow and overcome traditional bottlenecks, positioning it as the fifth paradigm of scientific inquiry, following experimentation, theory, computation, and data science. In recent years, AI4S has accelerated innovation across numerous scientific fields. The Nobel Prize-winning AlphaFold project has validated its substantial research value, prompting major global economies, including China, the US, Europe, and the UK, to elevate AI4S to a national strategic priority with a flurry of supportive policies. Concurrently, as general-purpose large models become more capable, model developers are seeking higher-value application scenarios to solve more critical problems. Scientific research, with its stringent technical demands and high returns on output value, has become a core target for these companies. While AI4S currently spans fields like life sciences, Earth sciences, and energy materials, its low penetration rate suggests significant headroom for growth, moving towards a phase of large-scale value creation.

AI4S Models Evolve Towards Integrated 'General-Specific' Architectures

Compared to general-purpose large models, AI4S models possess three exclusive core capabilities: scientific mechanism constraints, unified multimodal representation, and long-sequence reasoning. These features are essential for aligning with the real-world patterns of scientific research and tackling complex tasks. AI4S models are undergoing continuous iteration, transitioning from early-stage domain-specific and tool-enhanced models to a mainstream architecture that combines scientific foundation models with specialized domain models. Domestic institutions in China are making breakthroughs, launching integrated model architectures and intelligent research platforms like S1-Omni and SAGE. These platforms address the imbalance between a single model's generalizability and specialization, moving AI4S from isolated scientific computing towards a full-process, closed-loop scientific intelligence innovation system.

AI4S Demands a New Heterogeneous Architecture with Deep Integration of Supercomputing and Intelligent Computing

The computational characteristics of AI4S, which couple mechanistic simulation with AI models, impose entirely new requirements on computing power. It needs multi-precision mixed computation, long-duration stable operations, and a dual-ecosystem software stack, none of which can be satisfied by traditional supercomputers or general-purpose intelligent computing clusters alone. To address this, the industry is proposing a new heterogeneous architecture that deeply integrates supercomputing and intelligent computing to achieve full-spectrum adaptation in computational precision, hardware interconnectivity, storage, and software ecosystems. A leading domestic computing company, Orient Securities Company Limited, has launched the fully domestic, exascale-scale Sugon 8000 cluster. This platform serves as a self-developed, integrated AI4S computing foundation, enabling world-class scientific breakthroughs like trillion-atom simulations and accelerated protein folding, pushing domestic high-end computing from demonstration applications to large-scale replication.

AI4S Applications Flourish Across Sectors, Connecting Basic Research to Industrial Deployment

AI4S is empowering a wide range of fields, including biopharmaceuticals, materials and energy, industrial simulation, meteorology, and agricultural breeding. AI-driven drug discovery covers the entire pipeline from target discovery and molecular generation to clinical trials, significantly compressing R&D cycles and costs. In the industrial sector, AI4S is fostering a 'data + physics' dual-driven simulation paradigm, revolutionizing CAD/CAE/EDA toolchains. In meteorology, AI models are achieving operational forecasting capabilities, enhancing new energy power prediction. In agriculture, AI4S leverages multi-omics data analysis to accelerate the breeding cycle of superior crop varieties. By moving beyond its role as a mere tool, AI4S is driving industries to shift from trial-and-error experimentation to a prediction-first R&D model, establishing a complete chain from scientific discovery to engineering and industrialization.

Risk Warning

The report also highlights potential risks, including dynamic adjustments to regulatory policies, delays in technological development and commercialization, and intensifying competition within the sector.