By Peter Loftus
Eli Lilly Chief Executive Dave Ricks was on stage with Nvidia founder Jensen Huang earlier this year in San Francisco touting the company's tech prowess when Huang teased him about the painstaking process of developing new drugs.
"I'm really hoping that your industry moves from drug discovery which is kind of like wandering around the forest looking for truffles," Huang said, in front of a crowd of biotech and pharma investors.
Indeed, Ricks and the rest of the pharmaceutical industry are looking to expand beyond collecting soil samples and bark pieces to find new drugs and are instead turning their hopes -- and investment dollars -- to AI. Lilly first announced a partnership with chip-maker Nvidia in October to build what it called the industry's most powerful supercomputer, and expanded that in January with a $1 billion, five-year collaboration mixing their scientists and engineers in a new Bay Area lab aimed at discovering new medicines with AI tools.
They aren't alone. Rival Roche has already announced it is building an even bigger supercomputer in partnership with Nvidia. Companies such as GSK, AstraZeneca and Merck have announced billions of dollars worth of partnerships in recent months with tech and AI-focused biotech companies aimed at fully exploiting AI.
Drug companies have been talking about the potential for AI to supercharge drug development for years, but it hasn't materialized in a big way yet.
"There was this promise you'd see dramatic improvement" in the rate of success of drug clinical trials as a result of AI, said RBC Capital Markets analyst Trung Huynh. "I don't think that's happened yet. I don't think there's definitive proof that AI improves outcomes so far."
Part of the problem is that the amount of underlying scientific data to train AI models has been limited, and the cost of running high-volume computer experiments is high, though it is coming down, said Najat Khan, CEO of Recursion Pharmaceuticals.
Recursion is a pioneer in this area. The company was founded on the premise that it could train AI on cell images to better understand drivers of disease and therefore improve upon the 90% failure rate that currently weighs on drug development.
"The first wave, there's a lot of things that failed," Khan said. "The drug hunter mindset was missing for a long time."
Khan, who became CEO in January, said Recursion is making important progress. Its AI platform helped figure out that targeting a certain protein in the body was likely to help treat an inherited colorectal polyp disorder. Recursion used that finding to acquire an experimental treatment that hits that protein. In a small, early stage study, the company said its drug significantly reduced polyps in patients.
AI tools also helped speed up the time it took Recursion to design a new, experimental cancer drug to around 18 months, from an industry average of about four years. It could still be years before data proves the drug works, though, because studies in humans take time.
Breakthroughs haven't come quickly enough for investors, though. Last year, Recursion laid off 20% of its workforce after cutting back its research pipeline. It still hasn't brought an AI-enabled drug to market nearly 13 years after its inception.
Some companies are getting closer to proving AI's value in drug research. Japanese drugmaker Takeda has a psoriasis pill that it acquired from a company that used AI to discover it. That succeeded in big studies and the company is submitting it for U.S. regulatory approval this year. Another Japanese drugmaker, Astellas, used AI to perfect its experimental pancreatic cancer drug setidegrasib. That is in late-stage studies now.
Astellas CEO Naoki Okamura says AI can extend a drug's reach even after approval. The company is using AI to dictate which doctors company sales representatives should visit to promote the drug, as well as what to say in the visits.
All told, RBC Capital Markets estimates the technology could save the U.S. pharma industry about $90 billion over the next five years, boosting per-share profits by up to 13%.
Much of the AI payoff for drugmakers so far has been streamlining back-office tasks or speeding up manufacturing rather than making groundbreaking drug research advances. The advent of more advanced generative AI -- such as ChatGPT in 2022 -- opened up new possibilities.
AI companies have started marketing tools specifically to drug companies to help them work more quickly.
AlphaFold, developed by Google DeepMind, uses machine learning to predict protein structures that could aid in drug design. In April, OpenAI introduced GPT-Rosalind, which it says is intended to help researchers shorten drug-development timelines.
"We all changed our view," Pfizer CEO Albert Bourla said in an interview. "It has the potential to change everything we do, and I believe it will. From early discovery to late development, to creating better yields in manufacturing."
In manufacturing, AI-driven gains are already here.
"The reality is where we've seen all the benefits in AI so far is not actually in drug discovery -- it's actually in the rest of the process," said Diogo Rau, Eli Lilly's chief information and digital officer.
For example, Lilly executives thought they had maximized production of tirzepatide, the main ingredient in its top-selling weight loss drugs Mounjaro and Zepbound, to meet the explosive demand that led to widespread shortages early on.
To test if the company could make even more, Lilly last year created a "digital twin" to simulate the tirzepatide manufacturing process and then ran it through a machine-learning tool that it developed internally. The tool identified various combinations of factors such as pressure and temperature inside the drug's manufacturing equipment that could be tweaked to shorten the process, said Scot Lindsey, senior vice president of manufacturing and quality.
Lilly declined to disclose how much it shortened production times or how much more tirzepatide it was able to make, though Rau said the increase was "mind blowing for how many more patients we reached."
Write to Peter Loftus at Peter.Loftus@wsj.com
(END) Dow Jones Newswires
May 02, 2026 11:45 ET (15:45 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments