How Lilly Used AI To Crank Up Production Of Its Popular GLP-1s
How Lilly Used AI To Crank Up Production Of Its Popular GLP-1s
ByAmy Feldman,
Senior Editor.
Senior editor covering healthcare
Follow Author
Mar 07, 2026, 06:30am EST
Eli Lilly's Diogo Rau
Getty Images for Wired
Forget the drug discovery hype. Here’s how the world’s largest pharma company is seeing a payoff from AI right now.
There’s been huge hype around AI’s potential for drug discovery. But at Eli Lilly, the world’s largest drug company, the first big, unsung payoff from AI has been in manufacturing its popular GLP-1 drugs, Zepbound (for weight loss) and Mounjaro (for diabetes).
“We literally made more product last year than we possibly could have without AI,” Diogo Rau, Lilly’s chief information and digital officer, tells Forbes. While he declined to specify exact numbers, he says it was “enough that it would’ve been material in our earnings reports.”
That’s a big deal for Lilly because demand for these injectable drugs has been sky high — and the company has struggled to make enough of them. From late 2022 through 2024, the FDA determined there was a shortage of these drugs, which meant compounders were allowed to make them under certain conditions despite the drugs’ patent protections.
“That was top of mind for us that we want to not be on the shortage list,” says Rau, who joined Lilly in 2021 after a decade at Apple and reports directly to CEO David Ricks. “We had a process that we all thought we had optimized. The risk of being on shortage made us look [again] even though we thought we had a process that was as good as it could be.”
To crank up its GLP-1 production, Lilly used what’s known as a digital twin, a virtual representation of a factory that uses real-time data to show precisely what’s going on in the real-world, letting it test improvements in the digital world before rolling them out. Digital twins are increasingly used to optimize manufacturing.
Lilly was able to use artificial intelligence and its digital twin to make its manufacturing process more efficient — allowing it to produce the drugs in higher volumes than would otherwise have been possible. To do so, it modeled out everything about its factory from the machines to the inputs to the processes, allowing the digital twin to simulate different configurations to come up with the best option. “We thought, this looks too good to be true, but the physical world still matched the digital twin,” he says.
In addition, it was able to better detect defects in its injectors, Rau says. Its technology can, for example, take dozens of photographs of each one of those autoinjectors, from a variety of angles, in increments of a few hundred milliseconds, to monitor for any breakages.
It’s not nearly as sexy as using AI for drug discovery. But Zepbound and Mounjaro accounted for more than half of Lilly’s $65 billion in revenue last year. Its sales of Mounjaro reached $23 billion, double the $11.5 billion it reported in 2025, while its revenue from Zepbound surged to $13.5 billion from $4.9 billion the previous year. That growth helped propel Lilly to become the first healthcare company to hit a $1 trillion market cap late last year (it now trades just below that).
“We literally made more product last year than we possibly could have without AI.”
Diogo Rau, Lilly’s chief information and digital officer
Lilly is still using AI for drug development, but it’s a much longer game given how long it takes for a drug to go from idea to approval. In January, Lilly and Nvidia announced that they had teamed up to invest $1 billion in an innovation lab to tackle problems in the pharmaceutical industry, helped by a powerful supercomputer. That month, Lilly also signed a collaboration deal with hot AI startup Chai Discovery, which has raised $230 million at a valuation of $1.3 billion, to build an AI model that could accelerate discovery of biologic drugs. Biologic drugs are derived from natural sources such as proteins or cells, versus chemicals that are synthesized in a lab.
Any payoff in drug development remains far off. “That’s going to be mid-2030s, if not late-2030s, when those medicines are on the market,” Rau says. “It’s a big bet on the future.”
People ask, “How fast are we going to be able to make drugs now? Can we have drugs come out in six months or 18 months?,” he says. “That’s one thing that gets overhyped the most, and it has a critical risk of undermining AI in the industry because that should not be the expectation we have.”
MORE FROM FORBES
ForbesThis Startup Wants To Use Mini Robots To Treat Alzheimer'sBy Amy FeldmanForbesGene Editing Has Struggled To Go Commercial. This Nobel Laureate Has A $1 Billion Plan To Fix That.By Amy FeldmanForbesReed Jobs Lost His Father Steve To Cancer. Now His Cancer-Focused VC Firm Has Raised Another $200 Million.By Amy FeldmanForbesHow A Wall Street Analyst Started A $4 Billion Obesity Drug CompanyBy Amy FeldmanForbesWhy Yann LeCun’s Hot New AI Startup Is Targeting HealthcareBy Amy Feldman