Frost & Sullivan releases the '2025 China AI4LS Industry Development Blue Book' (with full text available for download)

From 'tool-assisted' to 'paradigm reconstruction', AI4LS reshapes the research rules of life sciences through the deep integration of machine intelligence and scientific principles.

Turing Award laureateJim GrayThe proposed four paradigms provide a classical framework for understanding the evolution of research models, which are currently being proposed by scientistsFifth Normal Form—— AI for Science (AI4S), an intelligent scientific research system centered on AI. The next-generation scientific paradigm represented by AI4S refers to the integration of artificial intelligence technologies such as machine learning and deep learning to analyze and process simulated and real data across multiple dimensions, modalities, and scenarios. Unlike traditional computers, its role is limited to replacing humans with complex and tedious calculations. AI pursues 'intelligence' more than anything else, hoping that computers can simulate human thinking, thereby enabling learning, reasoning, and decision-making.

Since the Renaissance, the evolution of scientific research has always proceeded along two parallel paradigms: data-driven and model-driven. The data-driven paradigm is bolstered by statistical methods and machine learning technology, but its inherent 'black box' characteristics make it difficult to explain the mechanisms behind conclusions. Model-driven research, on the other hand, promotes breakthroughs in theoretical science by revealing the underlying laws of nature, but it suffers from huge computational demands. When these principles are needed, people often find themselves in a situation where they have theories but cannot apply them.

AI4LS innovatively integrates the two deeply, that is, it distills empirical 'principles' from 'data', and can also use 'principles' to simulate and generate 'data'. Compared to the shallow modeling of symbol associations by large models in other fields such as natural language processing, the mechanism-level integration achieved by AI4LS makes complex system research possible.

Source: Analysis by Frost & Sullivan

From protein structure prediction in life sciences to new formulation design in materials science, and from climate modeling and energy system optimization in Earth sciences, AI4S is being integrated into multiple scientific fields. Some areas possess unique research characteristics and industrial needs, becoming the most thoroughly empowered and value-fully realized representatives of AI4S capabilities. Among them, life sciences are gradually establishing their position as one of the most ideal application scenarios for AI4S due to their deep data foundation, high complexity of problems, and broad application prospects.

The research subjects in the field of life sciences are highly complex and diverse, ranging from microscopic molecular structures to macroscopic physiological processes. Traditional methods often face many challenges such as massive data, difficult variable relationships to analyze, and long R&D cycles. AI4LS, however, has the natural advantage of processing multi-dimensional complex data, capable of quickly discovering potential patterns hidden behind vast amounts of data and transforming these patterns into precise prediction and optimization strategies, significantly improving the efficiency and success rate of life science research and industrial development. The Blue Book highlights the achievements of AI4LS, including AlphaFold 3, Chai-2, and Boltz-2.For more information, please refer to the full report.

Source: Analysis by Frost & Sullivan

When summarizing and predicting the development of AI4LS industry over a longer time period and from a more macroscopic perspective, based on the overall issues that the industry needs to solve, the history and foreseeable future development of AI4LS can be roughly divided into three stages:

• Concept Import Phase - 'Imitation' Stage (2016-2021): The role of AI4LS 1.0 is 'intelligent extension'. The algorithm basically relies on manually set rules for learning, belonging to the intelligent extension of data, and summarizes patterns through deep learning of massive experimental observations.

   

• Large-scale infrastructure construction period - "Prediction" phase (2021-2026): The keyword for AI4LS2.0 is "prediction," emphasizing boundaries, error evaluation, and verifiability. AI is no longer a "blind learner," but operates within a specific scientific framework with constraints and goals.

   

• Mature Application Phase - "Creation" Stage (2026 and Beyond): AI4LS3.0 has evolved from a "predictive assistant" to a "designer and explorer," possessing full closed-loop autonomous capabilities from functional requirements to structural design, truly realizing independent innovation in complex systems.

In addition, the Blue Book sorts out and presents typical applications and key turning points in each of the three major development stages of AI4LS.For more content, please refer to the full report.

Driven by industrial demand, escorted by policy guidance, supported by a technology matrix, and empowered by capital

● Experimental costs hinder scientific breakthrough

The continuous rise in experimental costs is becoming a rigid constraint on the transformation and upgrading of the life science industry. From early high-throughput screening to preclinical in vitro experiments, a large number of expensive and repetitive wet experiments are continuously driving up R&D costs, stretching the capital return cycle of new projects, and rapidly approaching the risk tolerance threshold for investors and enterprises. The shortage of professional technicians further magnifies the consumption of time and funds, causing frequent obstacles to innovation pipelines on their way forward. The Blue Book will disassemble experimental costs from the perspective of wet experiment procedures and repetitive manual operations.For more information, please refer to the full report.

●R&D cycle lengthening increases the risk of failure

In the life sciences industry, excessively long R&D cycles and low project success rates are core challenges that have long constrained industry efficiency and innovation vitality. Against the backdrop of capital prudence and intensified market competition, traditional R&D processes appear increasingly cumbersome and inefficient. Therefore, how to compress time costs, improve decision-making efficiency, and increase experimental hit rates while ensuring scientific rigor is becoming a key driving force for the accelerated integration of the industry towards AI4LS. The Blue Book will analyze the high risks associated with R&D cycles from two perspectives: linear processes and high failure rates, and clinical recruitment and design delays.For more information, please refer to the full report.

●Data scale limits research efficiency

In the process of life sciences moving towards a data-driven era, the increasingly massive data scale has become a key force driving industrial transformation. This trend is not only reflected in the surge in data generation speed but also includes obstacles to cross-institutional data integration and a scarcity of high-quality annotation resources, which comprehensively restricts research efficiency and knowledge transfer capabilities. The Blue Book sorts out the challenges brought about by data scale issues such as data explosion and processing pressure, data silos and sharing barriers, data quality and throughput bottlenecks.For more information, please refer to the full report.

●Ethical pressure affects experimental progress

Currently, experimental technologies represented by high-throughput screening can accelerate data collection, but their reliance on specialized consumables and large-scale animal models results in single experiments costing up to hundreds of thousands of dollars. Complex pharmacodynamic evaluations involving higher animals such as primates face implementation barriers due to escalating ethical review pressure. More seriously, the non-replicability of experimental protocols further exacerbates resource waste, resulting in a large number of initial experimental results that cannot be effectively transformed into clinical research stages. Resource-intensive R&D models drive up R&D costs and also pose systemic risks at the levels of ethical compliance and ecological sustainability.

From the perspective of policy trends in Europe and America, the FDA is accelerating the inclusion of emerging technologies such as AI and organ chips into drug evaluation processes through legislative and regulatory updates, gradually phasing out reliance on animal experiments. The EMA, with policy directives and scientific evaluations as its approach, is systematically promoting the widespread application of alternative methods to animal experiments in the development of chemicals and pharmaceuticals. The Blue Book provides a systematic review of relevant policies.For more information, please refer to the full report.

In response to AI4LS policies, the number of policies has been continuously climbing, with content becoming increasingly detailed. The focus has shifted from macro-level top-level design to demonstration applications and governance standards. Early policies mainly consisted of national-level plans and special funds, focusing on building big data platforms, high-throughput automated experimental centers, and public computing infrastructure; after entering the '14th Five-Year Plan' period, a series of application notices focusing on scenario demonstrations were issued successively, accompanied by solicitation drafts for generative AI management methods and standard construction, indicating that supervision and standardization are becoming the core of the healthy development of the industry. The State Council also issued the 'Opinions on Deeply Implementing the 'Artificial Intelligence+' Action' (Guo Fa [2025] No. 11) in August 2025, taking medicine and health as one of the application directions, and promoting the deep integration of artificial intelligence with biomedicine. At the same time, local governments are also continuously strengthening regional innovation ecosystems, focusing on demonstration zone construction, talent introduction and cultivation, incubation of specialized platforms, and support for the entire industrial chain.

Overall, the AI4LS policy not only increases funding and project support but also increasingly emphasizes the implementation of application scenarios, standard rules, and sustainable governance, laying a more comprehensive institutional foundation for AI-driven life science research and industrial transformation.

Source: Analysis by Frost & Sullivan

In addition, the Blue Book compiles AI-related industry development support policies introduced by national and local governments from 2017 to 2025.For more information, please refer to the full report.

● Data resources: The "infrastructure" in the field of life sciences

Large-scale, high-quality data resources are the cornerstone of AI4LS model training and scientific discovery, and the method of obtaining them is gradually shifting from traditional manual experiments to automated production models. The significance of automation lies in the exponential expansion of data output, which is more reflected through standardized, closed-looped, and highly credible data supply. The Blue Book provides an in-depth analysis of laboratory automation platforms defined as 'data factories' in life sciences.For more information, please refer to the full report.

In addition, current life science research exhibits the characteristic of coexistence of multimodal data. These data information collectively outline the entire panorama of life processes, far exceeding the insights that a single data dimension can provide. The exploitation of AI technology for its value lies in its ability to automatically extract high-dimensional features such as gene sequences, protein structures, imaging lesions, and clinical phenotypes, and to learn their relationships within a unified model, thereby revealing the laws of life science at a higher level.

●Algorithm Platform —— Extracting Scientific Discoveries from Massive Data

With the introduction of deep learning architectures such as Transformers and GNNs, AI models have significantly enhanced their capabilities in processing sequence information and multimodal inputs, making 'end-to-end modeling' from DNA sequences to protein structures, from compound structures to phenotypic behaviors possible. The Blue Book report summarizes examples from home and abroad including AlphaFold3, Lingo3DMol, Biotrajectory, and NanoForge.For more information, please refer to the full report.

Reinforcement learning optimization achieves dynamic optimization of key tasks such as experimental design, parameter tuning, and treatment pathway selection by constructing an interactive environment with experimental systems or physiological models. In addition, it simulates causal reasoning capabilities in complex processes and adapts to real-world scenarios. It is mainly applied in fields such as molecular synthesis pathway planning, biological experiment pathway search, and personalized treatment.

In addition, the Blue Book also analyzes the sub-modules of other technical matrices, including computing power platforms that provide large-scale and high-speed computing capabilities, domain knowledge embedding that ensures model outputs conform to scientific laws, and composite teams that support system operation and cross-innovation.For more information, please refer to the full report.

The Blue Book presents the investment and financing situation in the primary market of China's AI pharmaceutical sector from 2019 to 2024. Overall, the enthusiasm for investment and financing has cooled down after the peak. Against the backdrop of capital gradually returning to rationality and placing greater emphasis on technology implementation and industrial collaboration, the investment and financing logic in the AI4LS field is also undergoing profound changes.

Compared to the earlier single-focus on algorithm breakthroughs or conceptual innovation, current capital favors enterprises with structured capabilities—often possessing verifiable technical routes, scalable platform architectures, and practical application paths that meet industry needs. Capital is no longer betting on 'whether AI can be done,' but on 'whether AI can be embedded into the core processes of life sciences and continuously unleash value.' Under this trend, the Blue Book has summarized the characteristics of enterprises favored by capital from the dimensions of technology, application scenarios, commercialization paths, and industrial collaboration capabilities.For more information, see the full report.

AI4LS has widely penetrated into diverse application scenarios and begun to exert its effects. Challenges in technology, scientific integration, ethics and law, as well as ecosystems, urgently need to be addressed.

AI is reconstructing the technical paths and efficiency boundaries of key aspects of drug research and development, thereby enhancing design quality and development success rates.

●Reinventing the Core R&D Link of Pharmaceuticals

Artificial intelligence technology can construct stable and transferable prediction and generation models in low-sample, small-data, and complex non-linear problems, thereby simultaneously improving design quality and success probability at multiple stages. It has changed the passive mode of long-term experience-dependent and linear progress in drug research and development. From the perspective of pharmaceutical processes, AI technology has found suitable application scenarios in multiple links:

Source: Analysis by Frost & Sullivan

The Blue Book further analyzes AI in the detailed R&D processes including target discovery and validation, binding site confirmation, lead compound discovery, lead compound optimization, into resistance optimization, and preclinical validation, reflecting the systematic value of AI in reconstructing R&D paradigms.For more information, please refer to the full report.

● AI technology companies and drug pipelines

The AI drug pipeline is transitioning from early exploration to validation and confirmation stages, with key representative projects demonstrating feasibility and speed advantages in advancing from mechanism innovation to clinical implementation across multiple targets. The Blue Book sorts out the structural characteristics of AI-developed drug pipelines at different clinical stages, disease areas, and R&D phases.For more information, please refer to the full report.

●Representative AI companies in the drug R&D scenario

Different types of enterprises are conducting diverse explorations around platform construction, model-driven capabilities, and implementation. Representative companies are promoting the leap of AI from a tool to an empowering entity through differentiated technical paths and application models. The Blue Book lists four types of enterprises—AI+Biotech, AI+SaaS/technology platforms, AI+CRO, and AI+Hybrid—and their representative case studies.For more information, please refer to the full report.

Through systematic transformation and intelligent integration, the Smart Laboratory breaks through the efficiency bottlenecks and reproducibility challenges in traditional experimental systems, and constructs a new generation of life science R&D infrastructure centered on data-driven approaches.

● Analysis of pain points in traditional life science experiments

The Blue Book analyzes the key obstacles affecting the development of traditional life science experiments, including efficiency and throughput bottlenecks, data silos and management chaos, as well as deficiencies in reproducibility and consistency. It explains how these long-standing structural contradictions constrain the development efficiency, reliability, and intelligence level of life sciences, and are also the fundamental driving force behind the emergence of the new paradigm of smart laboratories.For more information, please refer to the full report.

● Development Goals of Smart Laboratories

The development goal of the Smart Laboratory is to comprehensively enhance the efficiency, standardization level, data value, and intelligent decision-making capabilities of life science research through systematic reconstruction. The Blue Book systematically elaborates on the development goals of the Smart Laboratory, focusing on four major dimensions: efficiency improvement, standardized reconstruction, data value mining, and intelligent decision-making drive.For more information, please refer to the full report.

● Top-level design framework guides the construction of smart laboratories

The construction of a smart laboratory is not achieved overnight but is advanced through an evolutionary process from basic to advanced levels, from local automation to overall intelligence. Each layer of the technical system undertakes and responds to corresponding development goals, constituting key pillars in the top-level architecture of the smart laboratory. The Blue Book presents the gradual construction and integration path of smart laboratories at five levels.For more information, please refer to the full report.

Source: Analysis by Frost & Sullivan

● Development path of laboratory automation

In the landscape of automation in life sciences, Chinese enterprises are rapidly expanding along four iterative steps: 'single module - workstation - system - intelligence'. In the early stages, the focus was on liberating labor through hardware and standardizing operational procedures. In the middle phase, system integration was used to achieve process continuity and data integration, ultimately relying on AI and algorithms to build a scientific research closed-loop with self-learning and optimization capabilities. The Blue Book systematically sorts out single module automation, workstation automation, system automation, and intelligent automation, as well as their representative products. Chinese enterprises are using these technical paths to upgrade smart laboratories from single-point equipment to system integration, constructing new scientific research infrastructure with autonomous execution and dynamic scheduling capabilities.For more information, please refer to the full report.

The Blue Book further expands the practical atlas of AI technology across more biotech chains—AI's precise design capabilities in synthetic biology scenarios, as well as its value in data interpretation in genomics and personalized medicine, are systematically sorted out and presented in the report.For more information, please refer to the full report.

Driven by industrial collaboration, cross-border integration, and independent innovation, China's AI4LS industry is accelerating the formation of global competitiveness and moving towards a transformation from technology following to paradigm leadership.

The Blue Book systematically sorts out the development trends of China's AI4LS industry in terms of industrial collaboration, cross-border integration, and independent innovation.For more details, please refer to the full report.

• Industrial cooperation promotes the commercial implementation of AI4LSChinese AI4LS enterprises are leveraging a deep industrial cooperation approach of 'verification + collaboration', 'standard + interface', and 'integration + implementation' to bridge the critical links from algorithm models to clinical applications, accelerating the transformation of technology implementation and the construction of business models.

   

• Cross-field collaboration drives the rise of platform-based enterprisesThe versatility of AI4S technology is driving Chinese enterprises to integrate across fields and replicate in multiple scenarios, with many platform-based companies rising rapidly and building diversified innovation ecosystems that connect industries such as medicine, materials, and energy.

   

• The nurturing environment driven by scientific research, industry, and innovationDriven by triple impetus of scientific research investment, industrial foundation, and innovation capabilities, Chinese AI4LS enterprises are gradually establishing systematic advantages to support pipeline development, independent innovation, and global competition, nurturing industry-leading enterprises for the international stage.

The Blue Book compiles case studies of AI4LS domain enterprises including Magnesium Gamma Technology, Wangshi Intelligence, Yulu Qianxing, and Jitai Technology. These enterprises are demonstrating a diversified technical path and vibrant industrial ecosystem in the field of AI4LS through technological breakthroughs and model innovations in key areas such as smart laboratories, drug research and development, molecular dynamics simulation, and nanodelivery.

Magnesium Robotics is a leading provider of autonomous intelligent agents in the field of robot technology application in China. It is dedicated to providing comprehensive autonomous intelligent agents and multi-agent solutions for smart laboratories and intelligent manufacturing scenarios. Through highly intelligent software and hardware systems, it promotes the industry's accelerated advancement towards an efficient, precise, and sustainable intelligent future. With its innovative 'Perception, Conception, and Execution' closed-loop technology architecture built in the fields of artificial intelligence and robot automation, Magnesium's multifunctional autonomous agents can independently execute complex tasks. At the same time, its modular product portfolio design allows for easy configuration and large-scale deployment in different scenarios, liberating manpower from repetitive work and enabling them to focus on high-value tasks in laboratory and manufacturing scenarios, laying a forward-looking layout for the development of the next generation of scientific research, manufacturing, and other fields.

Currently, Magnesium Giga is applying its technological advantages and product innovation capabilities to life sciences, chemicals, food and beverages, agriculture, consumer integrated circuits, new energy, and other industries. It continues to expand into more diversified strategic emerging fields, helping customers build intelligent organizations, break through technological boundaries, reshape operational paradigms, and become key drivers and best partners in leading industrial upgrading and transformation.

StoneWise was established in 2018 and is a technology company that uses artificial intelligence technology to drive new drug research and development. Leveraging breakthroughs in the underlying theories of AI drug research and development, deep governance of drug research data, accumulated knowledge in the industry, and strong software and engineering capabilities, StoneWise has built a multimodal AI 3D molecular generation large model base that can accurately generate molecules or molecular skeletons that fit the pocket structure of targets.

The company's molecular generation model is based on the GPT/Transformer framework and integrates algorithms such as geometric deep learning. The model can also serve as a base model. Partners can fully integrate their own data, cognition, and models on this basis for customized iterations. Currently, nearly a hundred pharmaceutical companies and research institutions are using it daily. Relying on this model, the company's internal self-developed pipeline has achieved remarkable results, with the fastest pipeline entering phase I clinical trials; at the same time, the entity database business derived from the base model is also actively expanding both domestically and internationally.

Suzhou Divamics Inc., a new drug early-stage research enterprise centered on artificial intelligence, quantum mechanics, and molecular simulation algorithms, has positioned itself to serve the global market since its inception. Relying on its self-developed Divamics multi-scale molecular dynamics platform and Biotrajectory kinetic trajectory database, the company adopts an original 'molecular movie' drug design concept to provide AI-driven Hit-to-PCC integrated drug R&D services for biopharmaceutical enterprises. Currently, it has served more than 50 pharmaceutical industry clients at home and abroad, collaborating to advance over 80 new drug R&D pipelines. The company's headquarters is located in Suzhou, with a computing center in Beijing and BD centers in Singapore and Japan.

Jiantai Technology is a biotechnology company driven by artificial intelligence innovation in nanomaterials, focusing on using targeted drug delivery and discovery technologies to accelerate patients' access to innovative therapies in critical disease areas. The company was co-founded by Dr. Chen Hongmin, an academician of the US National Academy of Engineering, along with Dr. Lai Caida and Dr. Wang Wenshou, MIT scientists, and has received important qualification certifications such as the national 'Little Giant' specialized and innovative enterprise and national high-tech enterprise.

The company has independently developed NanoForge, a platform that provides global-first artificial intelligence-driven nanodelivery solution offerings. It owns the world's largest tens of millions-level LNP lipid library and has built three core solutions based on NanoForge: AiLNP (AI nucleic acid delivery system design platform), AiRNA (AI mRNA sequence design platform), and AiTEM (AI small molecule formulation design platform). The company has achieved the ability to precisely deliver LNP to eight key organs and tissues, including the liver, lungs, muscles, and immune cells, making breakthrough progress in the challenging problem of multi-organ and multi-tissue targeted delivery. It provides opportunities for drug development for tumors, metabolic system diseases, autoimmune diseases, neurodegenerative diseases, etc., and also offers possibilities for life forms to resist aging at the organ level.

The 'Blue Book' introduces some companies in the AI4LS industry, including company profiles, business segment layouts, core product pipelines, competitive advantages, and technology platforms.