Why Big Pharma Is Betting Big On AI

In the mid-1960s, a small team at Stanford tried to teach a computer how to think like a chemist.

The project was called DENDRAL, and the task was simple to describe but brutal to do. It looked at messy mass-spectrometry data to figure out what molecule could plausibly have produced it.

Back then, chemistry was still more of an art than a science.  

You learned the tools, you watched the masters, you built instincts, and you got humbled by molecules that refused to behave.

DENDRAL treated the whole thing like a search problem.

It generated candidate structures, applied constraints, and eliminated what didn’t fit, until the remaining answers were not “pretty,” but probable.

That early idea matters in 2026 because drug discovery is still, at its core, a long sequence of guesses under uncertainty.

Let’s face it, most drug candidates fail. Many fail late in development, and failures cost companies billions of dollars… not to mention the time they wasted. 

So, the industry kept coming back to the same question: How do you get BETTER at guessing, quicker?

That’s where AI is quietly moving from curiosity to infrastructure.

Not as a robot scientist replacing the lab bench, mind you, but as a system that can rank hypotheses, predict binding, and propose new molecules. 

Today, AI is telling us what drugs are worth making next. 

Big Pharma Is Betting Big On AI

This month, a major pharma company inked a $1.7 billion, multi-year AI drug discovery partnership built around AI-driven small-molecule design.

When deals like that get done, you can bet it’s because someone has decided that enough is enough. 

Look, AI drug discovery works best when you stop thinking of it as a single model, and start thinking of it as a loop.

The loop starts with a target hypothesis, a structure (or a proxy for one), and a set of constraints that matter in the real world: potency, selectivity, safety signals, metabolism, and whether the molecule can even be synthesized efficiently.

From there, the modern systems do two things unusually well…

They search large chemical spaces quickly, and they learn from feedback without needing a human to hand-label every step.

The result isn’t certainty, but rather triage.

Why? Well, if you can rule out weak candidates sooner, you spend fewer quarters chasing dead ends.

Makes sense, right?

And if you can propose higher-quality starting points, you then compress the early discovery timeline that used to take half a decade to complete.

This is why the biggest signal right now isn’t a flashy demo! It’s the industry’s willingness to build around AI as a permanent capability, including compute and automated experimentation.

Do you think it’s a coincidence that Eli Lilly jumped in bed with NVIDIA to build a first-of-its-kind AI co-innovation lab that will apply advanced AI infrastructure to drug discovery and connect models to lab reality through robotics and scaled workflows. 

That’s the part most people miss.

Keep in mind that drug discovery doesn’t fail because teams can’t generate ideas.

Rather, the issue is that biology can be messy, and the distance between an idea and a validated candidate is bogged down with two very crucial elements — time and money. 

If AI is going to matter, it has to address these bottlenecks, not just do the brainstorming.

So what are the bottlenecks AI is already pressing on?

Well, just think…

• Target identification is shifting toward AI-guided biology, including large-scale perturbation and CRISPR-informed datasets.

• Structure and interaction modeling is becoming more accessible, including approaches that help predict how molecules may bind to proteins.

• Candidate generation is moving from “screen what exists” toward “design what should exist,” especially in small molecules.

• And clinical-trial measurement and consistency are starting to get formal AI support from regulators, which reduces friction for adoption.

That last point is worth lingering on, because it changes the tone of the whole conversation.

Last December, the FDA announced it had qualified its first AI drug development tool for use in clinical trials for MASH, aimed at standardizing and speeding the way disease activity is assessed from liver biopsy images.

But this isn’t a case of AI designing a miracle drug, dear reader. 

No, this is far more practical. It’s regulators drawing clear boundaries for how AI can be used to produce information that supports decisions about safety, effectiveness, and quality.

The FDA has also highlighted guidance work on “Considerations for the Use of Artificial Intelligence to Support Regulatory Decision Making for Drug and Biological Products,” which is another way of saying the agency expects more AI-generated evidence to show up, and it wants it to be auditable.

That matters because drug discovery is a chain.

AI Just Broke Drug Discovery Forever

There’s a familiar pattern we find within markets. 

That is, the loudest excitement gathers around the most visible outcome. 

In biotech, that outcome is a new drug approval, preferably one that changes a standard of care.

However, the economic value often starts accumulating earlier, in the unglamorous layer where the workflow becomes repeatable.

That’s where AI drug discovery is drifting toward right now. However, it’s not because models are perfect, but because the cost of running the old process is rising while the incentive to move faster is becoming structural.

You know as well as I do that drug pipelines are crowded, competition is global, and a single program costing billions can burn years of hard work only to arrive at a market that has already moved on.

So the winners in this era won’t necessarily be the people who talk the most about AI.

If you’re trying to think a step ahead, then consider what happens next…

As this becomes normal, certain companies won’t just “use AI.” 

Instead, they’ll own a meaningful piece of the discovery pipeline that others rent. 

Some will be platform-first specialists partnering with larger drugmakers, while others will be hybrid outfits that pair models with lab throughput, so they control both prediction and proof.

Then, there’ll be a small group of biotech players that sit on the most valuable choke points — the data, the compute architecture, or the translation layer that turns biology into a design space.

You can’t help but wonder which of those biotech gems will keep showing up on Wall Street’s radar. 

Well, how about I show you one right here.