The promise of AI in drug discovery is enormous: molecule design in hours instead of years, virtual screening of billions of compounds, clinical trial predictions that reduce failure rates. The reality, as of 2025, is more nuanced. AI is accelerating parts of the drug discovery pipeline, but it is not replacing it — and most AI-designed drugs are still years from pharmacy shelves.
DeepMind’s AlphaFold 3, released in 2024, represents the current state of the art in protein structure prediction. It can predict not just how a protein chain folds — the breakthrough that earned AlphaFold 2 recognition as Science’s 2021 Breakthrough of the Year — but how proteins interact with DNA, RNA, small molecules and other proteins. The database of predicted structures, freely available to researchers worldwide, contains over 200 million entries covering virtually every known protein. The tool has been cited in over 20,000 scientific papers and is used by more than two million researchers globally.
Insilico Medicine, a Hong Kong-based AI drug discovery company, has been one of the most visible examples of AI-designed molecules reaching clinical trials. Its lead candidate, a drug for idiopathic pulmonary fibrosis discovered using its Pharma.AI platform, entered Phase II clinical trials in 2023. The entire discovery process — from target identification to preclinical candidate nomination — took approximately 18 months and cost $2.6 million, compared with the industry average of 3-5 years and tens of millions of dollars.
But the distance between a Phase II trial and an FDA-approved drug is vast. Historically, about 30% of drugs that enter Phase II progress to Phase III, and only about 50% of Phase III drugs receive approval. The real test of AI in drug discovery will not be whether it can propose interesting molecules — that has been demonstrated — but whether those molecules actually work in humans. That evidence will take years to accumulate.
Recursion Pharmaceuticals, which went public in 2021, takes a different approach: using AI to analyse cellular images at massive scale. Its platform generates over 2 million images per week of cells treated with different compounds, using computer vision to detect subtle changes that indicate therapeutic effects. The company has partnerships with Roche-Genentech and Bayer, and its pipeline includes programmes in oncology, neuroscience and rare diseases. As of early 2025, Recursion had three drugs in clinical trials, with results expected over the next 2-3 years. The verdict on AI-driven drug discovery is still pending — but the data is accumulating faster than ever before.