insitro

insitro is a privately held drug developer in South San Francisco, founded in 2018 by Daphne Koller, the computer scientist who co-launched Coursera. It appears here not as an adopter bolting AI onto an existing business but as an AI-native drug developer: it pairs machine learning with biological data it generates at scale, from cell imaging to human genetics, to pick disease targets and then design medicines against them. It does not sell an AI product. Its platforms, which it calls Virtual Human, ChemML and TherML, run its own pipeline and reach other companies only through partnerships, and it now describes itself as a physical AI company.

Its defining pattern is data-generation-first. Rather than relying only on public datasets, insitro builds its own and trains models on them. In February 2026 it applied computer vision to MRI scans from 69,598 UK Biobank participants to estimate brown fat, a tissue that burns energy, and ran a genetic study on the result; the company says its lead brown-fat target cut body weight by 15 percent in obese mice while sparing lean mass. In January 2026 it agreed to acquire CombinAbleAI, an antibody design startup in Rehovot, Israel, and launched TherML to design small molecules, oligonucleotides and antibodies in one system.

Large drugmakers have paid for access to that approach. Gilead signed first, in April 2019, paying 15 million dollars upfront for a fatty liver disease collaboration covering five targets. Bristol Myers Squibb followed in 2020 with 50 million dollars upfront for amyotrophic lateral sclerosis (ALS), and up to 2 billion dollars in milestones. That deal has produced measured results rather than promises: 25 million dollars in milestones and a first ALS target in December 2024, an extension in October 2025 to design molecules with ChemML, and two more targets, plus 10 million dollars, in March 2026. Eli Lilly signed in October 2024, giving insitro an option on Lilly's liver delivery technology for two insitro small interfering RNA (siRNA) drugs. In September 2025 insitro began building models, trained on decades of Lilly's drug data, that predict how a molecule is absorbed, metabolized and cleared; they form part of Lilly TuneLab, open to biotechs that partner with it. The company says it has raised more than 750 million dollars, including a 400 million dollar Series C in March 2021.

The gap is the clinic. As of September 2026 no insitro medicine has been tested in people. Its most advanced program is a liver-targeted siRNA against a gene called IRS1 for metabolic dysfunction-associated steatohepatitis (MASH), which is in the safety studies required before a first human trial. In June 2026 it reported mouse data showing reduced fibrosis markers beyond the drop in liver fat, and in September 2026 it hired Hideo Makimura as chief medical officer to lead its move to a clinical-stage company.

The honest framing: insitro's value rests on whether targets found by its models hold up in patients, and with no drug yet in human trials, that remains untested. What is measured is that partners keep paying for targets, with Bristol Myers Squibb nominating two new ones in March 2026. AI is the whole method rather than a lever on top, which is also the risk. Recursion, which merged with Exscientia, built its own automated labs for cell imaging; Insilico Medicine already has its lead drug, rentosertib, in a Phase 3 trial; Isomorphic Labs and Xaira bet more heavily on protein structure and design.