Tech Giants Rush into AI Pharma: Experimental Data and Closed-Loop Capabilities Become the Competitive Focus

Since the beginning of 2026, the density of moves by global technology and pharmaceutical giants has been unprecedented: Nvidia and Eli Lilly announced a joint $1 billion investment to build an AI drug R&D laboratory in San Francisco, planning to deeply integrate computing power and model-building capabilities with Lilly's biological data over five years. Almost simultaneously, Mark Zuckerberg's Biohub, together with Google DeepMind, Meta, the U.S. Department of Energy, and the NIH, launched an $1.8 billion "Virtual Cell" initiative aimed at generating and standardizing large-scale biological data for AI. The common thrust of these events is clear: technology companies are no longer satisfied with "empowering" pharma; they are directly entering the field to restructure the underlying infrastructure of drug discovery.

However, what truly determines the outcome of this competition may not be computing power or model scale, but a more fundamental question: whether it is possible to continuously produce high-quality, iterable experimental data designed specifically for AI.

The trend of global technology companies accelerating their entry into biomedicine is becoming increasingly clear. Life sciences are becoming an important battlefield for tech giants competing for the next stage of AI capability boundaries, following large language models, autonomous driving, and robotics. Biomedicine may be the most valuable and also the most rigorous testing ground for validating next-generation AI capabilities.

But there is a fundamental difference in the pharmaceutical industry: data. Internet companies possess massive user behavior data that is generated at high speed daily; in biomedicine, truly valuable data is often expensive, scarce, and highly dependent on experimental conditions. A set of rigorously designed efficacy experimental data, or a set of strictly evaluated safety experimental data, is typically far more difficult to obtain than text that can be scraped at scale from the internet.

More and more technology companies are beginning to realize that the truly scarce resource in the next stage may be high-quality, iterable experimental data generated specifically for AI. Traditional AI training data largely comes from records left by humans in the past; life science data is highly dependent on experimental conditions, and data generated by different laboratories, equipment, batches, and processes is often difficult to directly concatenate. A future cycle may emerge: AI designs molecules → automated systems complete synthesis and experiments → activity and other properties are measured → results enter the database → the model is retrained → AI designs the next batch of molecules. Drug discovery thus has the opportunity to evolve from the traditional Design-Make-Test-Analyze process into a machine-learning-style closed-loop optimization, forming a continuous learning system that enables the model to continuously approach better solutions in the "design–experiment–feedback–iterate" cycle, driving a paradigm shift in drug discovery from experience-driven to data- and algorithm-driven.



From Experience-Driven to Data- and Algorithm-Driven

Entrogen Biotech and Youngen Biotech have established a deep strategic partnership, integrating their respective core strengths in wet-lab platforms and AI algorithms to co-build a dry-wet closed-loop system for small nucleic acid drug design, and based on this system, to develop an AI agent for integrated sequence and modification design. The sequences and modification schemes generated by this agent are experimentally validated by Youngen Biotech's wet-lab platform. Multi-dimensional data including activity, stability, off-target effects, and immunogenicity are structurally recorded and fed back into the database, based on which the model continuously iterates and improves in performance. In this way, Entrogen Biotech has built a continuous learning system of "AI design – wet-lab validation – data feedback – model evolution," enabling the integrated challenge of mutually constraining sequence and modification in small nucleic acid drug design to continuously learn and approach the complex mapping relationship of sequence–modification–activity within a vast combinatorial space. At present, the two parties have established a platform capability that can continuously produce high-quality experimental data and drive the operation of the data flywheel.