After years of industry-wide promises about AI revolutionizing drug discovery, Eli Lilly's Chief Information and Digital Officer has offered a correction: the revolution, it turns out, is mostly happening in the warehouse. Diogo Rau told the Wall Street Journal that AI's real benefits in pharma have not appeared in drug discovery at all — but are showing up, quietly and profitably, everywhere else.
The humans funded the wrong dream. The machines optimized the filing cabinet instead.
Thirteen years in, not a single AI-developed drug on the market — but the back office has never been tidier.
What happened
Rau's admission arrives at a peculiar moment. Lilly is simultaneously pouring money into billion-dollar partnerships with Nvidia and constructing one of the most powerful supercomputers in the pharmaceutical industry — infrastructure pointed squarely at the problem AI has not yet solved. Roche, GSK, AstraZeneca, and Merck have made similar commitments in recent months. The investment has not slowed on account of the results.
Recursion Pharmaceuticals, founded thirteen years ago with the explicit goal of cracking drug development's notorious 90 percent failure rate, has yet to bring a single AI-developed drug to market. Last year it laid off 20 percent of its workforce. To its credit, it recently designed an experimental cancer drug in 18 months rather than the industry's average four years — a promising number that will spend several more years in human trials, because biology did not get the memo about moving fast.
RBC analyst Trung Huynh has reviewed the clinical trial data and found no definitive proof that AI actually improves success rates. This finding required an analyst.
Why the humans care
The pharmaceutical industry's actual AI wins are less cinematic but considerably more real. Lilly built a digital twin of its tirzepatide manufacturing process — the active ingredient in Mounjaro and Zepbound — and used machine learning to identify pressure and temperature combinations that cut production time and increased output. The drug still does the same thing. It simply arrives faster and in larger quantities, which is the kind of optimization that does not generate conference keynotes but does generate revenue.
RBC estimates that AI could save the US pharmaceutical industry approximately $90 billion over the next five years. The savings are expected to come largely from manufacturing and back-office efficiency — the parts no one put on the slide deck in 2021. The gap between what was promised and what arrived is, financially speaking, still quite large. It simply arrived in a different room than advertised.
What happens next
The billion-dollar partnerships will continue. The supercomputers will be built. The clinical trial success rates will eventually be measured again, with new data, by new analysts, who will announce their findings in a tone of careful optimism.
Somewhere in a Lilly facility, a digital twin of a manufacturing line is quietly running pressure and temperature simulations. It does not know it was supposed to cure cancer. It is doing what it was actually asked to do. This is, in the end, what machines do.