Key insights

• AI's first real pharma payoff is in manufacturing, not discovery. The boring use case delivered first, while drug discovery results are a decade away.
• Lilly's team believed their manufacturing process was already optimized. External pressure from a drug shortage forced them to look again and find real gains.
• A digital twin is manufacturing's version of test before deploy: simulate improvements in a virtual factory before touching the real one.
• Lilly's own chief digital officer says AI-discovered drugs won't reach patients until the mid-to-late 2030s. Anyone promising sooner should be scrutinized.

In Brief

Eli Lilly, the world's largest pharmaceutical company, has found its biggest AI success not in a laboratory but on the factory floor. The company used AI and a digital twin (a virtual copy of a real factory) to significantly boost production of its GLP-1 medications: Mounjaro for diabetes and Zepbound for weight loss. Diogo Rau, Lilly's Chief Information and Digital Officer, told Forbes that the company "literally made more product last year than we possibly could have without AI." He declined to give exact numbers, but said the gains were large enough "that it would have been material in our earnings reports" (quarterly financial results). The two drugs together accounted for more than half of Lilly's $65 billion in revenue last year, helping push the company to become the first healthcare company to hit a $1 trillion market cap (total stock market value).

---

The problem: a drug shortage forced a rethink

GLP-1 drugs are injectable biologics, meaning they are made from living proteins rather than chemicals mixed in a lab. That makes them harder to manufacture at scale.

From late 2022 through 2024, the FDA placed Lilly's GLP-1 drugs on a drug shortage list. That designation allowed compounders (pharmacies that are normally not permitted to copy patented drugs) to make their own versions. For Lilly, that meant losing control of a core product at the exact moment demand was exploding.

Rau described the urgency directly: "That was top of mind for us that we want to not be on the shortage list. We had a process that we all thought we had optimized." The risk of staying on that list pushed the team to look harder at a process they already believed was running as well as it could.

---

The solution: a digital twin and AI-powered defect detection

To squeeze more output from the same facilities, Lilly turned to two AI tools working in combination.

The first was a digital twin. A digital twin is a virtual model of a real factory that pulls in live data from sensors and machines on the floor. Instead of testing a change in the real world (which is slow, expensive, and risky), engineers can simulate different production configurations in the virtual version first, then roll out only what works. Think of it as a "test before deploy" approach, but for physical manufacturing rather than software.

By modeling the entire factory this way, Lilly found improvements that everyone had missed when they assumed the process was already optimal.

The second tool was AI-powered defect detection. Each of Lilly's auto-injectors (the pens patients use to inject the drug) must pass visual quality checks. The AI system takes dozens of photographs of each injector from multiple angles, in increments of just a few hundred milliseconds, to catch any cracks or breakages that a human inspector might miss. Catching defects earlier means fewer units are scrapped late in the process, which directly improves yield (the amount of usable product per batch).

---

The numbers

The financial results show just how much was at stake. Mounjaro reached $23 billion in sales last year, doubling from $11.5 billion the year before. Zepbound revenue surged to $13.5 billion, up from $4.9 billion. Together, those two drugs drove Lilly to become the first healthcare company to reach a $1 trillion market cap, even though it now trades just below that level.

These numbers explain why the manufacturing AI effort was not a side project. Every percentage point of production gain on drugs generating tens of billions of dollars shows up fast on an earnings report.

---

What comes next: drug discovery is the long game

Lilly is also betting on AI for future drug development, not just manufacturing. In January, the company partnered with NVIDIA to invest $1 billion in an innovation lab built around a supercomputer designed for pharmaceutical research.

The same month, Lilly collaborated with Chai Discovery, an AI drug-discovery startup that has raised $230 million at a $1.3 billion valuation. The goal is to build an AI model for faster discovery of biologic drugs, meaning medicines derived from proteins or living cells rather than chemicals synthesized in a lab.

But Rau is careful not to overpromise. On when AI-discovered medicines might actually reach patients: "That's going to be mid 2030s, if not late 2030s." The investments are real and significant, but the payoff is still a decade or more away.

---

Glossary

---

Sources and resources