How Eli Lilly, NVIDIA, and Other Big Pharma Companies Use AI for Drug Discovery in 2026
The AI for drug discovery market hit $8.8 billion in 2026 and is projected to reach $114.4 billon by 2033.
Why? Because big pharma just found a way to cut drug development from 10-15 years to 6 months.
In January 2026, Eli Lilly and NVIDIA announced a $1 billion AI co-innovation lab to reinvent drug discovery. But Lilly isn’t alone. Pfizer, Novartis, Roche, Sanofi, and AstraZeneca are all doubling down on AI in drug discovery to slash costs and timelines.
Here’s exactly how the top AI drug discovery companies are doing it in 2026.
1. The Lab of 2026: Billions of Molecules Tested Before Breakfast
Let’s be real. For 50 years drug discovery was expensive guesswork. You’d pick a target, synthesize 500 compounds, and hope one didn’t kill a rat.
That’s dead now. The new workflow starts in a data center, not a lab.
Step 1: Find the real target. AI models now chew through genomics, proteomics, and 30 years of failed clinical trials overnight. They don’t look for correlations. They look for causation. Which protein actually drives Alzheimer’s? Which gene makes this cancer resistant?
Step 2: See the target. Thanks to AlphaFold 3, we get the protein’s 3D structure in 20 minutes. If you can’t see the lock, you can’t design the key.
Step 3: Invent the key. This is where it gets sci-fi. Generative AI like diffusion models and reinforcement learning don’t search databases. They dream up new molecules. From scratch. With the right shape, weight, and binding properties. It’s like having 10,000 medicinal chemists working in parallel.
Inside the $1B Lilly + NVIDIA Monster
Jensen Huang didn’t mince words in January: “We’re going to explore billions of possibilities in silico before a single experiment is run.”
What does that actually look like? Lilly runs a wet-lab experiment at 9am. The data hits NVIDIA’s BioNeMo platform by noon. The AI retrains. By 6pm it suggests 3 new molecules with better binding affinity. Lilly synthesizes them the next day. Loop. Repeat.
This isn’t theory. Lilly’s TuneLab is live. It’s trained on $1.2B worth of Lilly’s proprietary data and they’re already licensing it to smaller biotechs who could never afford that data.
The arms race is on. Sanofi bet on Exscientia. Pfizer on IBM Watson. Merck on Atomwise. Takeda on Numerate. If you’re a pharma CEO in 2026 and you don’t have an AI co-pilot, your board is asking questions.
2. Repurposing: The Cheat Code Big Pharma Loves
New drug = 15 years. Repurposed drug = 2-3 years. You do the math.
AI is a master at this because it can read every paper, patent, and side-effect report ever published and connect dots humans miss.
The poster child is still baricitinib. BenevolentAI flagged Lilly’s arthritis drug for COVID in early 2020. Lilly ran with it. It saved lives and made them $2B+.
What’s hot in 2026:
- J&J + BenevolentAI: Took an old antihistamine, bavisant, and pushed it for excessive sleepiness in Parkinson’s. Phase II data looks solid.
- Healx: Their AI found HLX-0201 for fragile X syndrome. 18 months from AI hit to Phase II. The old timeline was 5+ years.
- Cancer surprises: AI keeps flagging cheap generics like cimetidine and bazedoxifene for oncology. Same molecule, $10 vs $10,000.
For investors, this is the fastest path to revenue. No Phase I safety. Straight to Phase II efficacy.
3. Where The Billions Are Really Saved: Trials and Factories
Everyone obsesses over molecule design. The CFOs obsess over everything after.
Smarter Clinical Trials: ADMET kills 90% of drugs. Absorption, Distribution, Metabolism, Excretion, Toxicity. Pfizer is now using AI to predict drug-drug interactions by fusing molecular structure with EHR data. One McKinsey review put the time saved at 18-24 months per program.
AI Patient Selection: Instead of recruiting 2000 random patients, AI finds the 200 who are most likely to respond. Roche is doing this now for oncology. Smaller trials, faster results, higher success rates.
Digital Twin Factories: This is Lilly’s 2026 flex. Using NVIDIA Omniverse, they built a virtual copy of their manufacturing plant. They can simulate a raw material shortage, a machine failure, or a demand spike before it happens.
They’re also rolling out robotics for “physical AI” to scale GLP-1 drugs. Novartis uses AI to optimize formulations so pills dissolve better. Less waste, more yield.
The 2026 Scoreboard: Who’s Doing What
| Company | Core AI Play | Signature 2026 Move |
|---|---|---|
| Eli Lilly | Proprietary Data + Supercomputing | TuneLab platform, $1B NVIDIA lab, Baricitinib repurposing |
| Pfizer | ML for Discovery + Manufacturing | IBM Watson partnership, AI to boost vaccine yield |
| Novartis | Formulation + Multi-omics | AI-driven drug delivery optimization |
| Roche | Personalized Medicine | Patient-specific AI for treatment response |
| Sanofi, AZ, GSK, Merck | Partner with AI Biotechs | Deals with Exscientia, Insilico, Valo Health |
And don’t sleep on OpenAI. In April 2026 they dropped GPT-Rosalind, a foundation model trained only on chemistry and protein data. First customers: Amgen and Moderna.
My Take: The Moat Isn’t The Model. It’s The Data.
Here’s the uncomfortable truth for AI startups in 2026. Your algorithm is probably good enough. But you don’t have the data.
Eli Lilly has 100 years of wet-lab results. Failed experiments, toxicology reports, manufacturing data. That’s the moat. And that’s why the NVIDIA deal is a big deal. It’s the first time that kind of private data meets that kind of compute.
The other milestone to watch: FDA approval. As of today, not one drug has been 100% designed by AI and approved. When that happens in 2027 or 2028, the $114.4B forecast will be laughable. We’re talking $300B+ industry overnight.
the best data + the best AI + the fastest manufacturing. That’s exactly the bet Lilly and NVIDIA are making right now.
Final Thoughts: Why AI Drug Discovery Matters in 2026
AI hasn’t delivered an FDA-approved de novo drug yet. But it has already found life-saving repurposed drugs, cut timelines, and saved billions.
The race isn’t who discovers the next drug. It’s who builds the best AI for drug discovery to discover it first.
Want the technical breakdown of how AI models are trained on molecular data, ADMET prediction, and foundation models for chemistry?
