The $2.7B Question: How Do We Know If Isomorphic's AI Works?

Isomorphic Labs has raised about $2.7 billion from outside investors since 2025. Its president, Max Jaderberg, says its drug design engine works, but the company has not publicly named a drug candidate.

A new manifesto, and skepticism

On September 29, Isomorphic's Max Jaderberg published "Building a new path to make medicines with AI". "Our hypothesis has been proven," he writes, and the company's preclinical data is "helping us to gear up for clinical developments."

The centrepiece is a design run on an unnamed "Target D." Isomorphic's engine, IsoDDE, searched chemical space on its own for two to four days. A handful of molecules were then made and tested, and one produced a working concentration-response curve. Isomorphic compares that with a published molecule it estimates took three to five years of work.

Melinda B. Chu (@MelindaBChu1) was unimpressed. "Isomorphic is slower than traditional R&D and with $2 billion," she wrote. "No IND in 5 years. Not even real preclinical data reported; only vague Target D." Her prediction: no trial in 2026, and possibly not in 2027.

We replied that counting drug candidates might not be the only measure of AI success. She pushed back saying Isomorphic is a drug company, its founders talk about curing disease, and "you can't cure a disease without a drug." She also called our view "overly optimistic."

This piece sets out both cases and the public record behind each.

The skeptics case: big money, delays, little disclosure

Here is how Chu's points compare with the public record.

The money is real, and larger than she said. Isomorphic raised $600 million in its first outside round in 2025, then a $2.1 billion Series B in May 2026, led by Thrive Capital with Alphabet, GV, Temasek and others. That is about $2.7 billion from outside investors, on top of Alphabet's earlier funding.

The clinic keeps moving. Demis Hassabis had said Isomorphic would have AI-designed drugs in trials by the end of 2025. At Davos in January 2026, he pushed that to the end of 2026. In mid-September, head of drug discovery Chris Butler said the company was "on track" in preclinical work but declined to give a clinical timeline.

Almost nothing is disclosed. There is no named candidate, no named target and no program count. The only pipeline detail is two focus areas: oncology and immunology. Target D has a curve but no potency figure, no selectivity or ADMET data and no in vivo results.

Some AI peers have arguably moved faster. Chu’s own example points to Insilico Medicine. Indeed, it took rentosertib, its drug for idiopathic pulmonary fibrosis, from target discovery to phase 1 in under 30 months. It has since published phase 2a results in Nature Medicine and started phase 3.

As of September 30, 2026, Isomorphic has not announced an IND filing or a trial start.

The optimist case: partners, benchmarks and normal biology timelines

The case for Isomorphic rests on partnerships, benchmarks and timelines.

Big pharma keeps signing, and coming back. In January 2024, Isomorphic signed its first two deals. Eli Lilly paid $45 million upfront, with up to $1.7 billion in milestones; Novartis paid $37.5 million upfront, with up to $1.2 billion. A year later, Novartis added up to three more programs on the same terms. In January 2026, Johnson & Johnson joined with a multi-target deal covering several types of molecule; terms were not disclosed. Pharma companies that have seen the internal data are expanding, not leaving.

The engine has public numbers. In February 2026, Isomorphic published a technical report on IsoDDE. It claims more than double AlphaFold 3's accuracy on hard protein-ligand cases, better antibody-antigen prediction and binding-affinity estimates that match physics-based methods. These are benchmarks, not drugs, but they can be checked.

Five years without an IND is not unusual. The industry average from a first hit to a candidate molecule is 33 to 36 months, before IND-enabling safety studies even begin. Isomorphic also spent its early years building the platform, including AlphaFold 3 with Google DeepMind in 2024, rather than running programs from day one.

The company says it is on schedule. Its own public target is trials by the end of 2026. That has not been missed yet.

In our exchange, we pointed to CytoReason, which sells AI disease models to pharma and has never built a pipeline. Pfizer extended its deal with CytoReason in 2022 for up to $110 million over five years; the company has raised about $130 million in total. Chu replied that Isomorphic was never pitched as a software company, and that platform revenue would not come close to its valuation.

Schrödinger, which sells drug design software and also runs its own pipeline, gives a sense of scale. In 2025 it reported $199.5 million in software revenue and a $103 million net loss. Isomorphic does not sell its engine at all. Its income comes from partners, and the public figures are $82.5 million in upfront payments from Lilly and Novartis. We found no public report of a milestone payment, which would show a partner program reaching an agreed scientific goal.

Isomorphic has set its own test: a first clinical trial by the end of 2026, which is also a delay from its previous goal of late 2025. Chu thinks it will not happen this year, but ultimately time will tell, and even if it doesn’t, will it be a definitive indicator of the platform’s limitations in this regard? 

Go deeper into the increasingly lucrative and sophisticated techbio scene with BiopharmaTrend Pro Membership.

Topic: AI in Bio