The pharmaceutical industry is placing an increasingly large wager on artificial intelligence, betting that machine learning can compress the slowest and most expensive stage of making a new medicine.
According to a Forbes column by Tim Bajarin, published August 24, 2026, AI could cut preclinical drug research and development costs and timelines by as much as 70%. Preclinical work is the long stretch before a drug ever reaches a human volunteer — identifying promising molecules, screening them against biological targets, and weeding out candidates likely to fail. It is a phase historically defined by trial, error, and years of laboratory grind.
A 70% reduction, if it holds up in practice, would be a structural change rather than an incremental efficiency gain. Drug development economics are built around the assumption that most candidates fail expensively and late; anything that makes the early failures cheaper and faster changes what companies can afford to attempt in the first place, including treatments for smaller patient populations that never penetrated the business case before.
But Forbes frames the central question carefully, and it is worth repeating: the real test of AI-driven drug discovery is not speed or cost savings. It is whether the approach actually delivers safe, effective medicines to patients. Faster identification of drug candidates means little if those candidates do not survive clinical trials, where human safety and efficacy are established and where the majority of drug failures still occur.
That distinction — between generating more candidates and generating better ones — is where the industry's bet will ultimately be settled. The technology's promise is measurable in months and dollars today; its validation will only arrive in the form of approved therapies years from now.
Why it matters: if AI genuinely shortens the path from laboratory idea to approved treatment, it could lower the cost of medicines and put therapies within reach for diseases that were previously too expensive to pursue.