Anthropic and OpenAI Are Both Going After Drug Discovery. Here's Why That Matters.

Two things happened in the last two weeks that most people in the AI space aren't connecting.

Anthropic paid $400 million for Coefficient Bio — a drug discovery startup with fewer than 10 people that's only been around for eight months. The founders came from Genentech's internal computational drug discovery unit. Then today, OpenAI announced a partnership with Novo Nordisk — the company behind Ozempic — to integrate AI across their entire drug development pipeline, from research to manufacturing. Full rollout by the end of this year.

These aren't side projects. This is where the money is going.

And they're not alone. Eli Lilly signed a $2.75 billion deal with Insilico Medicine in March for AI-driven drug discovery across cancer, metabolic disease, and immunology. Nvidia has a billion-dollar partnership with Lilly. Google DeepMind's Isomorphic Labs is preparing for human trials. In the first three weeks of 2026, Lilly, GSK, and Pfizer each signed major AI platform deals.

Here's what I think is happening.

Right now, the AI industry is obsessed with coding. That makes sense. The biggest companies in the world need software, and AI that can write and debug code has an immediate, obvious market. But there's a deeper reason too — AI that gets better at coding can improve itself. Better coding ability means better AI research tools, which means better models, which means better coding ability. It's a flywheel.

But that flywheel doesn't need babysitting forever. Once the self-improvement loop is running in the background — and we're getting close — the AI labs need to point their capabilities at the next set of problems worth solving.

Science is the logical next step. And I mean that literally — it's logical.

Think about why AI is so good at coding. The feedback loop is binary. The code either runs or it doesn't. The test either passes or it fails. You can iterate on that endlessly because the outcome is measurable. Drug discovery works the same way. Did the molecule bind to the target? Did the compound reduce the tumor in trials? Did the treatment improve the biomarker? Yes or no. There are degrees of success, sure, but the core question is always binary — did it work or didn't it.

That's the kind of problem AI eats for breakfast. Once you know that something affects something else in a positive way, you can pursue that signal. You can iterate on the compound, tweak the dosage, test the variant. The same loop that makes AI great at debugging code makes it great at narrowing down drug candidates.

What Anthropic and OpenAI are doing now is moving from "here's an AI that can help researchers" to "we're building AI specifically designed to do the research." Anthropic didn't partner with a pharma company — they bought one. They're internalizing the domain expertise. OpenAI is embedding their technology directly into Novo Nordisk's operations. These aren't proof-of-concept pilot programs anymore.

And here's what really gets me excited. They haven't even scratched the surface yet. These are early days — first deals, first acquisitions, first integrations. The AI models being applied to drug discovery right now are general-purpose models being pointed at biology. Imagine what happens when they build models specifically trained for it. When the same kind of self-improvement flywheel that made AI great at coding starts spinning in biochemistry.

All it takes is one breakthrough. One discovery that extends the average human life expectancy by 50 years. One compound that reverses a disease we thought was permanent. That's not science fiction — the tools that could find it are being built right now, in deals that were signed this month.

I think five years from now we're going to look back at 2026 as the year AI labs stopped being just software companies and started becoming science companies. Coding was phase one. Drug discovery is phase two. After that? Energy. Climate. Materials science. The pattern is the same — find an industry where the problems are computationally complex, absurdly expensive, and slow, and throw AI at it.

The fact that both Anthropic and OpenAI made major pharma moves within two weeks of each other tells you neither of them wants to be second. And honestly? I've never been more optimistic about where this is all heading.

Sources: TechCrunch, CNBC, Bloomberg, PYMNTS, STAT News, Pharmaphorum