AI drug discovery deal: AstraZeneca pays insitro $40M upfront

A small check with a big ceiling

AstraZeneca has signed a multi-year deal with insitro, the California AI biotech, to look for new treatments for two hard-to-treat diseases: metabolic dysfunction-associated steatohepatitis (MASH), a serious liver disease, and cardiomyopathies, a group of diseases of the heart muscle.1 The British drugmaker will pay $40 million upfront. Insitro can also earn up to about $1.46 billion in development and commercial milestones, plus single-digit royalties on future sales.111

Outlets reported the size of the deal in different ways. Most rounded it to about $1.5 billion by adding the upfront payment to the milestones.614 The San Francisco Business Times called it worth up to $1.4 billion.15 Bloomberg's headline put the figure at $1.86 billion.17 The disclosed terms only add up to roughly $1.5 billion, and Bloomberg's own summary used the same $40 million and $1.46 billion figures as everyone else.17 The higher number may include some estimate of royalties, but the coverage doesn't explain it.

The more useful number is the split between guaranteed and contingent money. One trade publication worked out that the upfront payment is less than 3% of the headline value.5 Neither company disclosed how many targets are involved, the timelines or the payment schedule.5 My reading is that AstraZeneca is buying an option, not making a large commitment. It pays a modest fee to see what insitro's platform can find, and nearly all of the risk-adjusted value depends on that work producing real drug candidates.

How the work is divided

The deal sets out clearly who does what. Insitro will use its "Virtual Human" platform to find the biology that actually causes disease in both conditions. It will then take targets nominated by AstraZeneca from discovery to the development-candidate stage.11 After that, AstraZeneca runs and pays for clinical trials and keeps exclusive worldwide rights to the programs it picks.911 Insitro's own announcement adds that AstraZeneca will also handle commercialization.5 One trade outlet pointed out that insitro's role ends at the development candidate, so manufacturing, formulation and supply planning all fall to AstraZeneca.19

The two press releases describe the milestones slightly differently. AstraZeneca mentions development and commercial milestones, while insitro's version also includes discovery milestones.5 That matters more than it looks. If some payments are tied to discovery, insitro could get money in the early, platform-driven phase rather than only after a drug reaches the clinic.12

The reports also describe the platform a little differently. Reuters called it a mix of lab-grown cell models and machine learning.9 AstraZeneca described it as combining large-scale human and cellular data with AI tools for finding targets and designing medicines.11 Bloomberg added that insitro will use AstraZeneca's data as well as its own.17 If that's right, the deal is partly a data partnership. A large drugmaker's own data could make insitro's models better in ways that public datasets can't.

Why AI is being used to pick targets

The most notable thing about this deal is where in the pipeline AI is being applied. Much of the AI drug discovery story so far has been about designing molecules faster. This deal is about the step before that: deciding which biology to go after. AstraZeneca's R&D head, Sharon Barr, said choosing the right target is one of the most important decisions in R&D.11 She was more direct with Reuters, saying that by the time a new pathway shows up in a major journal it's no longer novel, and the company is already behind.1 Her argument is that spotting these connections early and testing them in cell systems puts AstraZeneca ahead.1

That's a competitive argument as much as a scientific one. Barr is saying AI is valuable because it can find proprietary, non-obvious targets before competitors read about them. She also said combining AI tools with experienced scientists means testing fewer molecules in the lab, less guesswork and a faster search for drugs that work.2 Notably, she presents AI as working alongside scientists, not replacing them.

The choice of diseases fits that argument. Insitro says both are heterogeneous: one diagnosis covers patients whose underlying biology differs, which has made treatment hard.5 Insitro's CEO, Daphne Koller, called both conditions common and underdiagnosed, and said detailed data on them holds a strong signal about what drives disease.11 Reuters reported that both have lacked effective treatments because researchers have struggled to find the root causes.1 Diseases where one label hides several different biologies are the kind of problem where pattern-finding in large human datasets could, in principle, beat traditional hypothesis-driven research.

MASH isn't untouched, though. The FDA gave accelerated approval to resmetirom (Rezdiffra) in March 2024 for adults with noncirrhotic disease and moderate to advanced liver scarring.5 The bet, then, is on finding more targets and different mechanisms, not on being first to treat the disease at all.

Where it fits for AstraZeneca

AstraZeneca's push into AI keeps growing. Fierce Biotech listed its earlier moves: a partnership with Algen Biotechnologies on immunology targets, the acquisition of Modella AI, and an agentic AI licensing deal with Owkin, alongside work on its own in-house tools.6 The insitro deal also fits AstraZeneca's focus on cardiovascular, renal and metabolic disease, which the company calls a key growth driver centered on links between the heart, kidneys, liver and pancreas.511 A program spanning a liver disease and heart-muscle diseases fits that framing well.

The market reaction was mildly positive. AZN shares were up 2.2% in premarket trading, though Stocktwits noted the stock had fallen more than 12% so far this year.7 A $40 million upfront payment is too small to matter to AstraZeneca's finances. The early move says more about investor mood than about how much the deal is worth.

What it means for insitro

For insitro, the deal matters more. The San Francisco Business Times called it much-needed for a South San Francisco company that went through painful layoffs last summer.15 Insitro says it has raised more than $750 million in capital and has partnerships with Bristol Myers Squibb, Eli Lilly and Gilead.5 Reuters noted that the company recently expanded its BMS partnership to cover more targets in ALS, and has worked with Lilly on AI models that predict how small molecules behave in the body.9

Fierce Biotech made a point other outlets missed: insitro's big-pharma deals are back-loaded, but they still bring in cash while it develops its own drugs.6 Insitro says it is advancing nine programs, including genetic regulators of liver fat and fibrosis that it expects to start clinical trials in early 2027.5 Insitro's release doesn't explain how those internal liver programs relate to the targets AstraZeneca will nominate.13 That overlap is worth watching. A company working on liver fibrosis biology for itself while hunting for MASH targets for a partner will need clear boundaries.

Overall, this looks like a familiar business model for AI biotechs. Partnership revenue pays for the platform and an in-house pipeline, and each new pharma partner signals that the technology is credible. AstraZeneca is now one of several large drugmakers using insitro's platform.6

What hasn't been proven

The trade press was more cautious than the companies. No candidate drug, efficacy data or safety data came with the announcement, and the work is still at the discovery stage.5 Targets found through AI analysis of human data still have to be validated in preclinical work and clinical trials, and the same variation between patients that motivates the AI approach could make those trials harder to design.13 Pharmaceutical Executive also noted that no independent check of the platform's performance in these diseases was available.5

Other coverage made the same point: the milestone money is contingent, and the deal's scientific value depends on whether AI-derived findings become validated targets and drugs that improve outcomes.1216 I agree. This deal doesn't show that AI can find better drug targets. It shows that a top-tier drugmaker is willing to pay a modest amount to test that idea in two diseases where conventional research has struggled.

The real test will come later, when insitro hands over its first development candidates and AstraZeneca decides whether to fund trials. If those programs move forward and the discovery milestones start paying out, the deal will be a strong example of AI picking targets. If they stall, $40 million will have been a cheap experiment for AstraZeneca, and the gap between the guaranteed money and the headline value will matter much more than the $1.5 billion figure.