Eli Lilly AI Strategy: Owning the Compute to Reinvent Drug Discovery

Lilly is betting that AI built on its own science, running on its own computers, will shorten the time it takes to make a medicine. It has pharma’s most powerful supercomputer, appointed its first Chief AI Officer in 2024 and shares its AI models with smaller biotechs. It has published almost nothing on what AI has produced so far.

The ambition is to break the pharma cycle. Diogo Rau, Executive Vice President and Chief Information and Digital Officer, said when the supercomputer went live in February 2026: “We’re trying to escape the traditional pharma industry life cycle” (Fierce Biotech). Fierce reported that he hopes AI can help cut the typical 10-year development timeline to five years.

AI is meant to be a scientist, not a tool. Chief AI Officer Thomas Fuchs said in October 2025: “Lilly is shifting from using AI as a tool to embracing it as a scientific collaborator” (R&D World).

Leadership is realistic about chatbots. CEO David Ricks said in March 2026 that general AI models asked to solve biology or chemistry questions are “not particularly good at it”, because “they’re trained on the human language, not on the language of chemistry, physics, and biology.” He said the future is “to build more and more models of those narrow prediction problems” (Fortune).

The focus starts in the lab. Lilly uses AI across:

• discovery: its own models for chemistry, biology and genomics, testing billions of molecule ideas on the computer before anything reaches the lab
• clinical development: trial design and patient enrollment
• manufacturing: a digital twin and computer vision to raise output of its GLP-1 drugs, Mounjaro and Zepbound, and catch defects
• the wider industry: TuneLab, which shares Lilly’s models with biotechs
• internal work: chatbots, agentic workflows and research lab agents built on internal platforms (February 2026)

Lilly has not said what it will not prioritize.

AI has its own executive. Thomas Fuchs became Lilly’s first Chief AI Officer in October 2024, joining from Mount Sinai. His remit covers discovery, clinical trials, manufacturing, commercial work and internal functions (pharmaphorum). He works alongside Rau, who has run IT, digital health, and advanced analytics and data science since 2021.

Lilly owns its compute. LillyPod went live in February 2026 with 1,016 Nvidia Blackwell Ultra GPUs. Lilly says it is the most powerful supercomputer in pharma, and it was built in four months (Nvidia).

When Fuchs joined in 2024, research computing was “very, very modest”: “eight GPUs for research, then 100, and now with the supercomputer.” He calls owning it “controlling our own destiny.” By June 2026, LillyPod was running large open-weight language models for the whole company with no token or budget limits (R&D World).

Partners and deals:

• Nvidia: LillyPod, plus a Bay Area co-innovation lab with up to $1 billion over five years (January 2026)
• Insilico Medicine: a deal worth up to $2.75 billion for drug candidates designed with generative AI (March 2026)
• OpenAI: using generative AI to find new antimicrobials (June 2024)
• TuneLab: Lilly’s own platform, launched in September 2025, gives biotechs access to models trained on data that cost more than $1 billion to generate. It uses federated learning, so neither side sees the other’s data (Lilly).

Everyone was told to use it. Business Insider reported in March 2025 that Rau “told everybody you need to use it, you need to start bringing ChatGPT into your work” (via NAM):

• In 2024, employees and managers were encouraged to use generative AI for year-end reviews.
• Lilly planned to require all senior leaders and managers to earn an AI certification in 2025.
• Lilly has not published completion figures.

The workforce stance is not stated. Lilly is growing and building new plants. It has not said how AI will change roles or headcount.

Governance is published and owned. Since 2024, a global “Using AI Responsibly” procedure has applied to all employees and contractors, owned by Rau as Chief Information and Digital Officer (Lilly):

• Eight principles guide AI use: accountable; safe; valid and reliable; fair and equitable; privacy-enabled and protected; transparent and explainable; sustainable; and enhanced.
• A cross-functional team reviews new AI uses, runs risk assessments and keeps a list of approved AI products.
• Bias audits are required for high-risk uses such as clinical work and HR.
• The board’s Audit Committee receives periodic briefings.

One line sums it up: “Employees are always responsible for the accuracy and appropriateness of all information generated using an AI system.” Manufacturing has the same instinct: “We don’t want any black box,” said Scot Lindsey, who runs IT for manufacturing and quality.

AI fluency, coded against the Lead with AI Fluency Matrix (September 2026 edition):

• Fluency floor: Level 1 (Chatting), set by expectation. Everyone was told to use ChatGPT, and managers were to be certified in 2025. Lilly has published no completion data or skill standard.
• Highest level evidenced: Level 3 (Integrating), company-claimed. Manufacturing teams use a digital twin and computer vision in production, with a reported effect on output. Lilly says teams build agentic workflows and lab agents, but it has published no results, so Level 4 is not evidenced.
• Portfolio design: Heavy on specialists. There is deep investment at the top (a Chief AI Officer, a supercomputer, its own models) and a broad usage push at the base. Little is published about everyday use in between.

Lilly reports capacity and deals, not AI outcomes. All figures below are company-claimed.

Manufacturing: Rau said “We literally made more product last year than we possibly could have without AI,” first reported by Forbes in March 2026. A digital twin produced a “tremendous reduction in the time that that bottleneck step took,” Lindsey said, though no figure was given (Pharma Manufacturing, June 2026). Finding that bottleneck depends on Workflow Literacy: knowing a process well enough to see where AI helps.

Discovery: One of Lilly’s own AI models helped design a promising drug candidate, and chemists are reusing the approach on other targets (June 2026). A lab team might test about 2,000 molecule ideas per target in a year. Now, “you can test billions of molecule ideas at your fingertips,” said Yue Wang Webster, who leads research and development informatics.

Results will take time. Rau said in October 2025: “for the medicines we’re working on now, we’re going to see those benefits in the 2030s at the earliest.”

Three things are missing: an AI value figure, employee adoption figures and AI targets. Lilly’s third-quarter 2026 results mentioned AI only in risk language.

The roadmap runs through the lab:

• the Bay Area co-innovation lab
• more of Lilly’s own AI models
• new models added to TuneLab
• renewable power for LillyPod by 2030

Change log

• 2021: Diogo Rau joins from Apple as Chief Information and Digital Officer.
• 2024: Lilly introduces its global “Using AI Responsibly” procedure. Employees are encouraged to use generative AI for year-end reviews.
• June 2024: An OpenAI collaboration on antimicrobials begins.
• October 2024: Thomas Fuchs is appointed Lilly’s first Chief AI Officer.
• March 2025: Business Insider reports that senior leaders and managers will need an AI certification.
• September 2025: TuneLab launches for biotechs.
• October 2025: Lilly and Nvidia announce pharma’s most powerful AI supercomputer.
• January 2026: Lilly and Nvidia commit up to $1 billion over five years to a co-innovation lab.
• February 2026: LillyPod goes live with 1,016 GPUs.
• March 2026: The Insilico deal, worth up to $2.75 billion, is signed, and Rau credits AI with higher GLP-1 output.
• June 2026: Company-wide large language models run with no token limits, and an AI-co-designed drug candidate is reported.
• October 2026: Third-quarter results mention AI only as a risk.

• Disclosure: 3 of 4. Lilly answers questions 01, 02 and 03 clearly. Measurement (04) is thin.
• It is the deepest AI infrastructure bet in the index. Lilly owns its compute and trains its own models.
• Its leaders set realistic expectations. The CEO says chatbots are poor at biology, and the CIO puts the benefits in the 2030s.
• The numbers reflect its drugs, not yet its AI. Revenue per employee rose 78% to $1.30 million from 2022 to 2025, currently the highest in the index, driven mostly by Mounjaro and Zepbound, while headcount grew 28%.
• Governance is published and owned: eight principles, review of every new AI use and board oversight.
• The say-do gap is the workforce. Every employee has AI with no token limits, but Lilly has not published how much people use it, whether managers finished certification or what it saved. That is a textbook capability overhang.
• Watch the Bay Area lab and the Insilico pipeline. The test is AI-designed drugs reaching clinical trials, and a number for the manufacturing gains.

What the numbers show: Revenue per employee rose 78% to $1.30 million, currently the highest in the index. Most of the growth comes from Mounjaro and Zepbound, and headcount still grew 28%. Lilly itself expects AI’s benefits in new medicines to arrive in the 2030s at the earliest.

Revenue, headcount and revenue per employee, FY2022 to FY2025

• FY2022: $28.54bn revenue, 39,000 employees, $732k per employee
• FY2023: $34.12bn revenue, 43,000 employees, $794k per employee
• FY2024: $45.04bn revenue, 47,000 employees, $958k per employee
• FY2025: $65.18bn revenue, 50,000 employees, $1.30M per employee

Change, FY2022 to FY2025: revenue up 128% (2.3x), headcount up 28%, revenue per employee up 78%.

Notes: Revenue is as reported. Headcount is approximate, at year-end.

These figures give context on scale and productivity. They do not show that AI caused any change. Sources: Eli Lilly 10-K filings.

What is Eli Lilly’s AI strategy?

Lilly uses AI built on its own research data to speed up drug discovery, clinical trials and manufacturing. It owns pharma’s most powerful supercomputer, trains its own models and shares some of them with biotechs through TuneLab.

‍

Who leads AI at Eli Lilly?

Thomas Fuchs has been Chief AI Officer since October 2024. He works alongside Diogo Rau, Chief Information and Digital Officer, who also owns Lilly’s responsible AI procedure.

‍

What is LillyPod?

LillyPod is Lilly’s AI supercomputer, built with Nvidia and live since February 2026. It has 1,016 Blackwell Ultra GPUs and runs discovery models and company-wide AI tools.

‍

How does Lilly govern AI?

Since 2024, Lilly has required employees to follow a “Using AI Responsibly” procedure based on eight principles. A cross-functional team reviews new AI uses, and the board’s Audit Committee oversees AI risk.

‍

How is this profile put together?

Lead with AI profiles use public sources only, such as annual reports, earnings materials, filings and on-record interviews, dated at the time of writing. AI fluency is coded against the Lead with AI Fluency Matrix. Company figures are marked as company-claimed unless they have been independently verified.