Drug discovery has gotten so efficient that the number of candidates in the pipeline has roughly doubled over the past decade, while novel FDA approvals have stayed flat at about 50 per year. That gap shows discovery isn’t the real choke point—clinical development is. Preclinical assets routinely fetch tens of millions in upfront licensing deals, but after Phase 2 proof of concept their value jumps into the hundreds of millions or even low-billions. As AI tools drive discovery costs down and crank out more molecules, that preclinical premium will shrink. Right now you’ll see seven or eight drugs vying for a single biological target; on blockbuster targets like PD-1 or GLP-1, more than 100 programs are already live. If discovery tools keep improving, expect that number to double or triple by 2030. When dozens of companies can generate chemically distinct candidates against the same pathway, each individual molecule becomes less rare—and investors will demand stronger translational data, smarter patient selection, tight dose optimization, and accelerated clinical paths before they write big checks. Improving discovery models is just one piece. AI-driven predictions work best where data is abundant and feedback loops are short—things like virtual screening, structural biology, even protein folding. They’ll struggle longer on messy problems with high biological variability: choosing endpoints, predicting human immune responses, running adaptive trials. The winners will be those who put their bets on the areas still bound by trial timelines and patient recruitment, rather than the areas that AI can commoditize in months.