AI in Drug Discovery: Which Parts of the Pipeline It Actually Speeds Up

When companies claim a drug was discovered using AI, they are typically referring to early-stage processes such as virtual screening, generative chemistry, or property prediction — not the clinical trials that dominate both time and cost. The full drug development pipeline spans over a decade, with clinical stages accounting for the bulk of spending and the highest rate of failure. AI tools can accelerate hit-finding and lead optimisation, potentially cutting years from early discovery, but these are the shortest and cheapest phases of the process. A key structural limitation is that property prediction models are trained on historical assay data, making them least reliable on novel chemical scaffolds that generative models are most likely to produce. Target identification, where AI could theoretically have the greatest impact, remains the weakest link due to the long feedback loop required to validate any nominated target.