The Zhitong Finance App learned that Orient Securities released a research report saying that AI4S currently covers fields such as life science, earth science, energy materials, etc., but the industry penetration rate is low, and it is expected to enter a new stage of large-scale value monetization in the future. AI4S is empowering multiple disciplines and industries, including biopharmaceuticals, material energy, industrial simulation, meteorological environment, agricultural breeding, etc. The rapid development and iterative progress of AI4S has brought investment opportunities on the scientific model side, computing power infrastructure construction and operation, and industrial application side.
Orient Securities's main views are as follows:
AI4S has become the fifth paradigm of scientific research and the “next stop” for big model iteration
AI4S refers to a new model that uses artificial intelligence to assist or independently complete scientific research tasks. It can reshape the scientific research system throughout the process, break through traditional scientific research bottlenecks, and is regarded as the fifth paradigm of scientific research after experiments, theory, computation, and data science. In recent years, AI4S has driven the acceleration of innovation in many fields of scientific research, and AlphaFold's winning the Nobel Prize also confirms its hard-core research value. As a result, major global economies such as China, the US, Europe, and the UK have upgraded AI4S to national strategies and intensively introduced policies to strengthen their layout. On the other hand, with the continuous enhancement of GM's large model capabilities, finding higher-value implementation scenarios and solving more important problems has also become the development direction of model companies. Scientific research scenarios have become the core choice of model companies due to their stringent technical capability requirements and high return on output value. At present, AI4S has covered fields such as life science, earth science, energy materials, etc., but the industry penetration rate is low, and it is expected to enter a new stage of large-scale value monetization in the future.
The AI4S model continues to iterate, and the single scene model is moving towards a comprehensive integrated architecture of communication and specialization
Compared with general models, the AI4S model has three exclusive core capabilities: scientific mechanism constraints, multi-modal unified representation, and long-term time series reasoning, so it can be more adapted to real-world scientific research rules and complex scientific tasks. The AI4S model is undergoing continuous iteration, gradually evolving from an early field-specific model and a tool-enhanced model to a mainstream general-specific integrated architecture with a basic scientific model plus a field-specific model. At present, domestic institutions continue to break through and launch integrated model architectures such as S1-Omni and SAGE and supporting intelligent research platforms, which solve the problem of the imbalance between generality and professionalism of a single model, and promote AI4S from single-point scientific computing to full-process, closed-loop scientific intelligence innovation.
AI4S requires a new heterogeneous computing power architecture with deep integration of supercomputation and intelligent computing
AI4S has computational characteristics that combine mechanism simulation with AI models to put forward new requirements for computing power. It requires multi-precision mixed computation, long-term stable operation, and dual ecosystem software stacks, which cannot be satisfied with traditional supercomputing and general intelligent computing clusters. Therefore, the industry proposed a new heterogeneous computing power architecture with deep integration of supercomputing+intelligent computing to achieve all-round adaptation of computing power accuracy, hardware interconnection, storage, and software ecology. Zhongke Shuguang, a leading company in the domestic computing power industry, launched a nationally produced 100,000 card Shuguang 8,000 cluster to create a fully self-developed supercomputing-intelligent computing AI4S computing power base, and relied on the platform to achieve world-class scientific breakthroughs such as trillion atomic simulation and accelerated protein folding, and promoted domestic high-end computing power from demonstration applications to large-scale replication.
AI4S has blossomed in many fields, opening up a closed loop from basic research to industrial application
AI4S is empowering multiple disciplines and industries, including biopharmaceuticals, material energy, industrial simulation, meteorological environment, agricultural breeding, etc. AI pharmaceuticals cover target mining, molecular generation, and even the entire clinical trial process, significantly reducing R&D cycles and costs; the industrial field spawns a “data+physics” dual-drive simulation paradigm, innovating the CAD/CAE/EDA tool chain; AI models in the meteorological field enable commercial forecasting to enable new energy power prediction; agricultural breeding relies on multi-omics data analysis to help shorten the breeding cycle of good species selection. AI4S gradually breaks out of tool-assisted positioning, driving various industries from trial and error experiments to a new R&D model that prioritizes prediction, and opens up a complete link from scientific discovery to engineering industrialization.
Risk Alerts
Regulatory policies dynamically adjust risks; technology development and transformation fall short of expectations; racetrack competition heightens risks.