Pharmaceuticals and Artificial Intelligence: From Faster Design to Better Odds, and What Remains Unproven

Abstract

Purpose. Drug development is long, costly and mostly unsuccessful, so any tool that improves the choice of targets, molecules or trial designs is valuable. This article asks where artificial intelligence stands in the pharmaceutical industry in 2026 and where it could be by about 2035. Design and methods. A review that combines the share of the drug-discovery and development literature that involves AI (OpenAlex, 2010–2025), Eurostat's survey of AI use in pharmaceutical manufacturing firms in the European Union (2021–2025), and published evidence from structural biology, clinical pipelines and regulators. Claims from the literature were checked against the cited sources. The scenarios are judgements built on this evidence. Findings. The share of drug-development works that mention AI rose from 1.2% in 2018 to 8.0% in 2025 and language-model works from 68 to 1,640 in three years. In the EU, 41.3% of pharmaceutical-manufacturing enterprises used AI in 2025, against 15.7% in 2021 and 19.9% of all enterprises, but the uses were led by text mining (25.6%) and language generation (21.0%), not machine learning for data analysis (12.6%). The scientific successes are real, from protein structure prediction to a generative-AI-designed molecule that showed a signal in a 71-patient phase 2a trial, and AI-discovered molecules passed phase I in 80–90% of cases; but phase II success, about 40% in a small sample, was similar to historic averages. We expect AI to remain routine in structure-based design and to move into trial design and development planning by 2028, and the first approval of an AI-designed drug to be a question for about 2029–2031. Originality. The article separates the parts of the pharmaceutical value chain where AI has demonstrably worked from those where it is still hoped for, uses a dated adoption series for the industry, and gives signposts for the main open question: whether AI improves the odds of clinical success or only the speed of design.