$115 million in hand

On March 29, Eli Lilly took an exclusive worldwide license to a portfolio of oral drug candidates that Insilico Medicine designed with generative AI — picking the targets and drawing the molecules by software. The headline number was $2.75 billion, one of the largest AI-drug-discovery deals yet. The fine print is that $115 million of it is paid now; the remaining $2.6 billion is owed only if the candidates clear development, regulators, and the market — milestones almost none of them have reached.

The cheap half is solved; no AI-discovered drug has yet cleared the FDA.

That structure is the honest measure of where AI in drug discovery actually stands. The software has compressed the cheap front half of the work — Insilico says it can reach a preclinical candidate in roughly a year for a couple of million dollars, against several years and far more by hand. (The widely repeated claim that it is '100 times faster and cheaper' appears in none of the deal's primary sources and overstates even Insilico's own figures.) What AI has not touched is the expensive back half: the human trials, where the timeline runs a decade-plus and roughly nine in ten candidates still fail. No fully AI-discovered drug has been approved by the FDA.

What makes Insilico worth a bet rather than a press release is one data point its rivals can't show. Rentosertib, the first drug whose target and molecule were both found by generative AI, posted positive mid-stage results in lung-scarring patients last year — published in Nature Medicine, not a slide deck. It is the field's first piece of human evidence that a machine-designed drug can work. The Lilly money says the front half is real; whether the back half follows is what the next decade pays out.