Genentech Partners With AI Antibody Firm Earendil on Cancer Bispecifics in Deal Worth Over $1.5 Billion

What happened

Earlier this week, Genentech, the Roche unit, agreed to collaborate with Earendil Labs on bispecific antibody therapies for cancer. The agreement was announced on Thursday, Sept. 24, and BioSpace reported that Genentech will pay $55 million upfront, with total potential value exceeding $1.5 billion including development, regulatory and sales milestones.

Earendil could also receive tiered royalties on net sales of products that reach the market.

Deal details

• Upfront: $55 million.
• Total potential value: More than $1.5 billion.
• Royalties: Tiered royalties on net sales.
• Scope: Several bispecific antibody programs in oncology; the target diseases have not been named.
• Roles: Earendil leads antibody discovery and early clinical research for pre-agreed target combinations; Genentech handles global clinical development and commercialization.

Company details

Earendil uses an AI-driven platform to generate antibody sequences for next-generation biologics. BioSpace reported that the company raised $787 million in a private round in March 2026 backed by Sanofi and the Biotech Development Fund, created by Hillhouse and Pfizer, among others.

The company has an existing pharma relationship. In April 2025, Sanofi paid $125 million to license two Earendil bispecifics for autoimmune and inflammatory bowel diseases, with up to $1.72 billion in milestones.

A Roche spokesperson said the company continues “to follow the science and invest in emerging breakthrough areas and assets with best-in-class potential.”

Why it matters for the sector

Bispecific antibodies, which bind two targets at once, have become a major focus in oncology. Roche already sells two CD20xCD3 T cell engagers, Lunsumio and Columvi, for lymphoma, and the Earendil deal could deepen that portfolio, according to BioSpace. Designing bispecifics is technically demanding because the two binding arms must work together without causing excessive toxicity or manufacturing problems, which is one reason developers are turning to computational design tools.

The deal is one of several large AI-related drug discovery agreements announced in late September. Large drugmakers have been signing collaborations with AI-focused companies that can generate candidates more quickly, often with modest upfront payments and milestones that add up to headline values above $1 billion.

Earendil’s financing history also illustrates the scale of capital now available to well-regarded AI biotech companies. A private round of nearly $800 million earlier this year, followed by partnerships with Sanofi and Genentech, gives the company resources to run its own early clinical work rather than relying solely on partners.

For Roche, the agreement adds external innovation to its oncology pipeline at a time when the company has been active in licensing, including deals for cardiometabolic and obesity assets.

The division of labor, with the platform company taking candidates into early clinical studies, also differs from many discovery deals in which the pharma partner takes over at the preclinical stage. That arrangement can let Earendil retain more value as programs advance.

The structure of the payments, with a relatively small upfront amount and most value contingent on success, is in line with other AI-enabled antibody deals. Investors often look past the headline figure to upfront cash and the number of programs covered when comparing such agreements.

Partnerships with two large drugmakers in different therapeutic areas also diversify Earendil’s revenue sources, reducing dependence on any single collaboration.

What to watch

• Disclosure of the target combinations and cancer types covered by the collaboration.
• Progress of Earendil’s partnered programs with Sanofi in immunology.
• Further Roche and Genentech dealmaking in bispecific and multispecific antibodies.
• Whether Earendil considers a public listing given its funding scale and partnerships.

Sources: BioSpace

Why Bispecific Antibodies Are Hard to Build

A conventional antibody is a Y-shaped molecule made of two identical heavy chains and two identical light chains, with both arms recognizing the same target. A bispecific needs two different arms, and that creates engineering challenges that do not arise with ordinary antibodies.

• Chain pairing. When cells produce two different heavy chains and two different light chains, they can combine in many wrong ways. Developers use engineered formats to make the correct pairing dominate.
• Geometry. The distance and angle between the two binding arms affect whether the molecule can bring two cells or two proteins together effectively.
• Affinity balance. Binding too tightly to one target and too weakly to the other can reduce effectiveness or increase side effects.
• Developability. Molecules must be stable, resist clumping, be manufacturable at scale and stay in the body long enough to work.

T cell engagers in particular

Many oncology bispecifics are T cell engagers, with one arm binding a tumor target and the other binding CD3 on T cells. They can be highly potent, but strong immune activation can cause cytokine release syndrome, a systemic inflammatory reaction. Dosing schedules that step up gradually are often used to manage that risk.

Where AI Antibody Design Fits

Traditional antibody discovery often begins with immunizing animals or screening large libraries, then refining hits through many rounds of laboratory work. Computational platforms aim to compress that process. In general terms, such platforms:

1. Learn patterns from large datasets of antibody sequences and structures.
2. Generate new candidate sequences predicted to bind a chosen target.
3. Score candidates for properties such as stability, expression and potential immune reactions.
4. Send the most promising designs for laboratory testing, feeding results back to improve the models.

For bispecifics, where many combinations of arms and formats are possible, the ability to explore more designs quickly is a central part of the pitch. Laboratory and clinical data remain the ultimate test.

Questions Readers Ask

What is a “pre-agreed target combination”?

Pairs of targets that the partners select in advance for each bispecific program. Defining them up front clarifies which programs fall under the agreement.

Why would Earendil run early clinical work?

Taking candidates into the clinic before handing them off can increase the value the platform company captures, since milestones and royalties often rise as risk falls.

What does CD20xCD3 mean?

It describes a bispecific that binds CD20, a protein on B cells including many lymphoma cells, and CD3 on T cells, bringing them together so T cells can attack the cancer.

How long until products could reach patients?

Discovery-stage antibody programs typically take many years to move through trials and regulatory review, and many do not reach the market. Patients should discuss treatment options with their own oncologists.

This article is for general information only and is not investment or medical advice.