CITIC Construction Investment: Building a closed loop of dry and wet data barriers, macromolecules are expected to raise the industrial ceiling

The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that AI4S is moving from a model capability competition to a tipping point of industrialization. Technology, industry, demand, and policies resonate four times, and the value anchor is being upgraded from model ability to scientific research productivity, data assets, and systematic biological design. Scientific research agents are upgraded from answering questions to disassembling goals, calling professional models, and connecting experiments to push AI from single-point tools to scientific research productivity to perform complex R&D tasks. Pay-as-you-go billing models such as Science Token open up platform-based monetization space; model generation is faster than experimental verification capabilities, and experimental production capacity and feedback cycles are the main constraints on R&D efficiency. Those who master real data entrants use dry and wet closed-loop data flywheels. The focus shifts from model performance to experimental throughput, data quality and feedback efficiency; AI capabilities shift from small molecules to antibodies Expanding macromolecules such as proteins and nucleic acids, and upgrading research and development from generating a single candidate to a systematic drug design that integrates pharmacogenicity optimization, delivery, and in vivo verification, is expected to open up industrial value space.

CITIC Construction Investment's main views are as follows:

Quadruple resonance drives AI4S to a critical point of industrialization. On the technical side, generative biological design expanded from single-molecule prediction to complete genomes; on the industrial side, general-purpose model manufacturers extended to wet experiments, and MNC formed horizontal alliances to build R&D infrastructure; on the demand side, candidate supply amplified synthesis and high-throughput verification requirements, and upstream service providers took the lead in implementing them; on the policy side, AI pharmaceuticals, high-quality data sets, and scientific research infrastructure were incorporated into the systematic layout, forming a synergy of technology, capital, orders and policies. The industrial value anchor is being upgraded from a single model capability to scientific research productivity, data assets, and systematic biological design, which is mainly reflected in three aspects:

(1) Paradigm evolution: Agents extend the R&D chain, and TaaS opens up value space. Agents are being upgraded from answering questions to disassembling scientific research goals, calling professional models, and connecting real experiments, so that algorithms, data, and devices form a continuous workflow. Science Token packages models, computing power, task orchestration, and robot experiments as standardized services. Customers pay per call volume, and the platform can transform reusable scientific research capabilities into continuous, high-frequency platform revenue.

(2) Deepening of barriers: dry and wet closed-loop construction of data flywheels, wet laboratories deposit data assets. The speed of model generation candidates is faster than experimental verification capabilities, and experimental production capacity, cost, and feedback cycle have become the main constraints on R&D efficiency. Technology giants master the real data production portal by building their own wet laboratories and connecting to CRO and automated test platforms; experimental results continue to flow back to training to form a “design-verification-feedback-iteration” self-reinforcing flywheel. The competitive focus also shifts from model performance to experimental throughput, data quality, and feedback efficiency. Automated wet laboratories are expected to become an important scenario for the implementation of physical AI.

(3) Border expansion: Moving from “molecular selection” to “manufacturing systems”, macromolecules are expected to raise the industrial ceiling. Small molecules are still the most mature battleground for data foundations, modeling tools, and clinical verification; the development of protein structure prediction, protein language models, and generative biology is driving antibodies, proteins, and nucleic acid drugs towards predictive-driven and programmable design. Due to the high degree of coupling between sequence, structure, pharmacogenicity, delivery, and in-vivo effects, macromolecular AI will upgrade from generating a single candidate to a systematic drug design with multi-objective collaborative optimization.