Schrödinger's zasocitinib cleared two pivotal Phase 3 plaque psoriasis trials in December, partnered to Takeda, with the NDA filing window queued for fiscal 2026 and the first regulatory approval of a generative-AI-designed drug now sitting on the calendar. Insilico's rentosertib hit its Phase 2a endpoints in idiopathic pulmonary fibrosis and published in Nature Medicine in June. 173 AI-originated clinical programs sit in the queue, fifteen to twenty entering pivotal trials this year. AstraZeneca's TRAVERSE in treatment-naive mantle cell lymphoma is the first FDA real-time trial streaming endpoints into the agency over cloud, twenty to forty percent of trial calendar collapsing on the new infrastructure.

The skepticism of the last decade ran on one objection. Pharma's 90 percent Phase 2 failure rate has always been the body refusing molecules that satisfied every chemistry criterion the bench could write down. A generative model trained on chemistry, the argument went, would inherit the same wall, because the model knows compounds and biology eats compounds. Two clinical readouts and a Nature publication later, the assumption that AI-discovered molecules die in the clinic at incumbent rates is held by no analyst still pricing the pipeline seriously.

The first FDA-cleared AI drug arrives the same fiscal year the agency lets a generative model stream endpoints into the regulator in real time. The discovery layer and the trial layer compress on the same calendar. The next decade of drug development runs on a stack the FDA validated this quarter.