Guotai Haitong released a research report stating that on the large model front, scientific large models are entering the core R&D processes of pharmaceutical companies. The firm believes that data generation and validation capabilities centered on the "dry-wet closed loop" are becoming the core competitive barrier in AI drug discovery, with wet lab capacity and high-quality data now scarce strategic assets; driven by the resonance of large models, pharmaceutical companies, and service providers, industry prosperity and R&D outsourcing demand are expected to continue rising.
Main viewpoints of Guotai Haitong are as follows
On the large model front, scientific large models are entering the core R&D processes of pharmaceutical companies. Overseas models are upgrading from "tools" to "agents plus wet labs": Anthropic established a wet laboratory in the San Francisco Bay Area and used approximately 950 Claude agents working in parallel for 21 hours to autonomously discover a previously uncharacterized bacteriophage enzyme system (ART); OpenAI led an investment in biosecurity company RedQueen Bio to enter AI biodefense; Isomorphic Labs' IsoDDE engine, in agent form, completed functional molecule design within 2-4 days. Domestic players, while iterating model capabilities, are building proprietary data moats in niche sectors: Insilico Medicine previewed the full deployment of its "agent layer" and MCP, XtalPi upgraded the XtalPi Science platform, and MGI Tech partnered with BioMap to co-build a cellular large model.
On the pharmaceutical company front, MNCs are accelerating AIDD capability building through "in-house development plus collaboration," with ecosystem-type partnerships landing intensively. Overseas layouts can be grouped into three categories: models entering the main R&D process, co-building platform wet lab network ecosystems, and collaborating on specialized models to accelerate development in specific domains. For example, Novo Nordisk introduced ClaudeScience into its R&D process, Eli Lilly partnered with Twist, Ginkgo, and GenScript to co-build the TuneLab data closed loop, and Roche announced "AI autonomous controllability" and disclosed that 40% of pipeline decisions already have traceable AI/computational contributions. The domestic deal list continues to lengthen: the nine BD collaboration agreements disclosed by Insilico Medicine during the year have a potential total value of approximately US$7.3 billion, XtalPi's collaboration with DoveTree carries potential milestones of up to US$5.89 billion, and the orally administered GLP-1 small molecule designed by AI from Drug Farm reached a commercialization collaboration worth up to RMB 1.75 billion.
On the wet lab front, construction is upgrading from "outsourced services" to "strategic infrastructure," and the global capacity "arms race" is accelerating. The firm judges that the faster the AI closed loop turns, the stronger the reliance on wet lab throughput, and lab demand growth is expected to outpace the growth in the number of AI drug projects. Overseas, Roche is building its own AI laboratory, Anthropic is setting up a biological laboratory, Eli Lilly and Nvidia are co-building a joint innovation laboratory with up to US$1 billion over five years, and Recursion maintains approximately 2.2 million wet experiments per week in throughput and 65 PB of proprietary data; domestically, GenScript raised approximately HK$2.33 billion in net proceeds from a placement, with about 70% directed toward AIDD wet lab capacity, XtalPi operates an automated laboratory cluster of over 10,000 square meters, BioMap and Harbour BioMed are co-building a dry-wet closed-loop laboratory, and Insilico Medicine released the LabClaw intelligent laboratory system, with capacity expansion and data infrastructure being ramped up simultaneously.
Risk warnings: Model iteration and clinical progress falling short of expectations, collaboration implementation and commercialization realization falling short of expectations, intensifying industry competition, and changes in data compliance and regulatory policies.