The headline says $2.5 billion. The cheque, for now, is about $18 million. Both numbers are real, and the gap between them is the most useful thing to understand about the wave of AI-drug-discovery deals making news — including the one announced this week.

At the BIO 2026 convention, the AI-drug-discovery company Insilico Medicine and the Korean pharmaceutical firm SK Biopharmaceuticals unveiled a collaboration to design new medicines for neuroimmune disorders of the central nervous system — conditions like neuroinflammatory and neurodegenerative disease. The press release led, as these always do, with the big number: a deal "worth up to $2.5 billion." It's a genuinely interesting deal. But that figure deserves an asterisk, and learning to read it tells you a lot about where AI-in-medicine actually is.

The real deal

Strip away the headline and here's what was actually agreed. Insilico will use its in-house AI platform — it calls it Pharma.AI — to discover and optimise drug candidates for a set of hard neurological targets. SK Biopharmaceuticals will then take the promising ones through the long, expensive business of late-stage clinical trials and, if they work, to market.

This is a real division of labour, and it reflects something true about the field: AI has become genuinely good at the early part of drug discovery — proposing molecules, predicting how they'll behave, narrowing millions of possibilities to a handful worth testing. It's one of the more substantive places AI is delivering, not just promising. Insilico has candidates from this kind of work already in human trials, which is more than most "AI for drug discovery" stories can say.

The $2.5 billion asterisk

Now the number. In pharmaceutical deals, the headline value is almost never money that changes hands. It's "biobucks" — the sum of every possible payment if everything goes perfectly, across the entire life of a drug that mostly doesn't exist yet.

Read the fine print of this one and it breaks down like this. Insilico receives up to roughly $18 million upfront and in near-term milestones — that's the real, near-certain money. The rest of the $2.5 billion is contingent: development milestones (paid only if a candidate advances through each trial phase), regulatory milestones (paid only if it gets approved), commercial milestones (paid only if it sells well), plus single-digit royalties on any eventual sales.

The catch is that most drug candidates never reach most of those milestones. The large majority fail somewhere in clinical trials — that's true of all drug discovery, AI-assisted or not. So the $2.5 billion isn't a valuation of the deal; it's the theoretical ceiling, the number you'd reach only if a still-hypothetical drug ran the entire gauntlet from lab to pharmacy without stumbling. The honest description is "up to $2.5 billion in milestones," and the word that matters most is "up to."

Why this matters

None of this makes the deal hype. AI-designed drug candidates attracting real partnerships with established pharmaceutical companies is genuine validation — a few years ago this work was a research curiosity, and now a major drugmaker is betting its development pipeline on it. That's the real story, and it's a good one.

But the eye-watering numbers attached to these announcements measure optionality, not cash — the value of a lottery ticket if it wins, not the price paid for it. As AI-drug-discovery deals multiply, you'll see a lot of "billion-dollar" headlines, and almost all of them will work this way. The useful habit is to look for two figures: the upfront payment (what someone is actually willing to pay today) and the total "potential" value (what they'd pay across a perfect future that usually doesn't arrive).

AI is genuinely changing how the first stage of drug discovery works. Whether it changes how many drugs actually make it to patients — the part that happens long after the AI's job is done — is the question those headline numbers quietly skip. Read the asterisk.