Are AI-powered pharmaceutical companies rising?

For example, Eli Lilly has invested in computational infrastructure and partnered with Chai Discovery to develop an exclusive, custom AI model. AstraZeneca’s self-developed AI agent system, AIDA, is expected to shorten the CMC development cycle by 50%. In early 2026, AstraZeneca acquired Modella AI and plans to fully integrate Modella’s foundation models and AI agents into its oncology R&D ecosystem—particularly to support clinical development and biomarker discovery. At the data level, beyond its vast legacy drug R&D datasets, Eli Lilly has also brought in BigHat Biosciences to rapidly generate diverse, high-quality datasets using its platform, feeding Eli Lilly’s internal foundational models. Furthermore, Eli Lilly actively pursues acquisitions of high-potential external pipeline assets—specifically AI-discovered drugs with best-in-class (BIC) or first-in-class (FIC) potential. Overall, traditional pharmaceutical companies are becoming key participants in the AI-driven drug discovery industry. However, the most significant impetus for the AI-pharma sector—especially in areas such as AI-powered drug discovery—comes from the rise of ‘AI-native’ companies: startups founded from day one on computational logic and AI model architectures—the so-called AI-pharma companies. Early AI-pharma firms like Schrödinger focused on physics-based molecular docking—for instance, Schrödinger’s core technology is a physics-first-principles molecular simulation platform, and its FEP+ (Free Energy Perturbation Plus) tool remains among the most accurate methods available for predicting binding free energy. Later, AI-pharma companies such as Atomwise attempted to predict drug activity purely through ‘dry lab’ approaches (i.e., computation and algorithms), but faced severe challenges in clinical translation due to the complexity of biological systems and noise in public databases. The core bottleneck in AI-pharma lies in data—and AI-pharma companies like Recursion have moved beyond passive reliance on external data by building automated wet labs that tightly integrate dry and wet experimentation. Their continuously operating wet labs generate large volumes of high-quality, standardized, proprietary datasets to train AI models; in turn, those AI models guide high-throughput wet experiments, which yield new data to further refine the models—creating a self-reinforcing, closed-loop ‘data flywheel.’ The convergence of generative AI and automated laboratory systems is driving a paradigm shift in drug discovery—one that may ultimately realize the vision of fully autonomous, end-to-end drug development pipelines: ‘prompt-to-drug.’ According to ACS Central Science, the ideal workflow of such integrated AI-driven drug discovery systems involves AI models performing target identification and drug discovery, followed by synthesis and wet-lab validation, selection of clinical candidates, and even generation of clinical trial protocols and post-marketing surveillance plans (Alex Zhavoronkov et al., 2026).

Although AI-pharma companies frequently secure massive funding, why does clinical implementation remain so challenging?

Although AI-pharma companies frequently secure massive funding, clinical implementation remains difficult—primarily because AI can significantly improve the efficiency of ‘finding a drug,’ yet it has not solved the fundamental biomedical challenge of ‘proving that a drug is both effective and safe.’ Many investors focus on AI’s immense potential in areas such as molecular design, target discovery, and protein structure prediction—but what truly determines a pharmaceutical company’s value is not how many candidate molecules it generates, but whether it can successfully advance candidates through animal studies and human clinical trials to achieve regulatory approval. This process remains highly uncertain. AI excels in early-stage R&D—not late-stage clinical development. Traditional drug development broadly consists of several phases: 1. Identifying disease mechanisms and therapeutic targets 2. Designing candidate molecules 3. In vitro experimental validation 4. Animal testing 5. Phase I clinical trials (safety assessment) 6. Phase II/III clinical trials (efficacy evaluation) 7. Regulatory review and approval Currently, AI’s most mature applications are concentrated in the earlier stages. Examples include: ・ Predicting drug–protein binding modes based on protein structures ・ Screening candidate molecules from massive compound libraries ・ Generating novel molecular structures ・ Optimizing drug-like properties These tasks involve relatively standardized data and are well-suited for machine learning. But once development reaches human trials, problems suddenly become far more complex: ・ Human metabolism may differ substantially from model predictions ・ Results from animal studies often fail to translate to humans ・ Genetic variability across patients with the same disease is enormous ・ A drug may show efficacy but carry unacceptable side effects These issues cannot be resolved simply by increasing computing power. Biological systems are vastly more complex than AI models. The logic of many AI-pharma startups resembles the following: ‘Scientists used to take years to screen molecules; AI can do it in months—so R&D speed will dramatically increase.’ This reasoning is not entirely wrong—but it overlooks a critical distinction: Discovering a potentially active molecule and proving it can become an approved drug represent two entirely different levels of difficulty. For instance, an AI model may predict that a given molecule attacks cancer cells—but once administered to humans, it might: ・ Fail to reach the target tissue ・ Be ineffective at tolerable doses ・ Cause toxicity when dose is increased ・ Trigger tumor resistance ・ Elicit unforeseen immune responses Living systems are not fully computable engineering systems. Aircraft design can be rigorously validated through extensive simulations—but drugs operate within a dynamic, highly variable biological environment. Phase II trials constitute the true stress test for AI-pharma. Many AI-pharma programs successfully enter Phase I, because Phase I primarily asks: ‘Can humans tolerate this drug?’ But Phase II begins addressing the pivotal question: ‘Does this drug actually work?’ This stage sees extremely high attrition rates. Some studies indicate that although numerous AI-related drug programs have entered clinical development, most remain in early-phase trials; very few have advanced to later-stage clinical validation. The reason is simple: Phase I success only demonstrates safety.

Zealand Pharma, Generate Strike AI Obesity Drug Deal Worth Up to $1.2 Billion

15 hours ago

Zealand Pharma and Generate Biomedicines have partnered on an AI-driven protein-based obesity treatment program worth up to $1.2 billion. Generate will lead research while Zealand oversees clinical development and commercialization following a $30 million upfront payment.

reuters

2026-06-07

European Stocks Drop on Mideast Escalation; Zealand Pharma Sinks

2026-10-13

Zealand Pharma, Generate Strike AI Obesity Drug Deal Worth Up to $1.2 Billion

Zealand Pharma, Generate Strike AI Obesity Drug Deal Worth Up to $1.2 Billion

FAQ