Biotech Industry · global
AI Drug Discovery Investment Doubles as Big Pharma Calls for Proof Behind the Promises
From experience analyzing 10,000 compounds to screening diseases using 100 million patient records, pharmaceutical executives outline AI’s practical role: it can accelerate research, but it cannot yet replace evidence of efficacy and safety in humans.
A long process of validation still separates designing a seemingly promising molecule from turning it into a drug patients can use. As funding for AI drug discovery grows rapidly, executives at several major pharmaceutical companies have recently issued public reminders: computational speed can change the pace of research, but it does not make the complexity of human biology disappear.
A report published by Chosunbiz on November 10 and a Wall Street Journal report republished by Mint on November 4 both highlight this gap between investment enthusiasm and the realities of drug development. Both cite McKinsey data showing that global venture capital and private equity investment in AI drug discovery reached $8.4 billion in 2025, more than double the approximately $4 billion invested two years earlier. The accelerating flow of capital has made an industry question more urgent: can generating drug candidates faster lead to more successful therapies?
Vertex Executive Chairman Jeff Leiden offered a concrete example. According to The Wall Street Journal, about 50 AI companies had analyzed a Vertex dataset containing 10,000 compounds, but he said none had produced a usable drug as a result. However, the report did not specify each company’s task, evaluation timeframe, or criteria for success. This corporate experience cannot be treated directly as a failure rate for the entire field, but it reveals the distance between data analysis and drug development outcomes.
That gap also includes safety. Beyond acting on their intended targets, candidate molecules may affect other proteins or physiological pathways, causing unpredictable side effects. AstraZeneca CEO Pascal Soriot and Regeneron Chief Scientific Officer George Yancopoulos shared a caution: much remains unknown about disease mechanisms, and advances in models alone do not justify concluding that cancer will be cured or that development timelines will shrink substantially. These statements reflect the judgments of research and development executives, rather than a new clinical study comparing AI with traditional methods.
Clearer uses emerge in narrowing the scope of research and supporting decisions. The Wall Street Journal reported that a Bayer team used AI to analyze 100 million patient records within days, selecting five diseases to prioritize for research using its gene-editing technology based on genetic markers and medical histories. This work demonstrates efficiency in organizing large volumes of data; however, ranking diseases does not itself establish that the technology can be safely delivered to target tissues or that patients will receive therapeutic benefits.
Background and Context
The question of how to evaluate AI in drug development is expanding from whether models can identify molecules to whether they improve the entire research and development process. Identifying potential targets, refining candidate molecules, and helping find suitable trial participants all have practical uses, but each use requires different evidence. Strong predictive performance by a model cannot be taken directly as evidence of a higher clinical success rate, and one company’s experience is insufficient to draw conclusions about every platform.
These statements from pharmaceutical executives therefore set more concrete requirements for demonstrating results as investment continues to heat up: can research teams show, under comparable conditions, that AI reduces unproductive experiments, improves the quality of drug candidates, and carries those advantages through to human trials? The reports currently provide corporate experiences and application examples, but consistent comparative data remain lacking. The transition from research assistant to reliable drug development capability still hinges on reproducible, testable results.