Roche's Genentech Pays $55M Upfront for Earendil's AI-Designed Bispecifics

Genentech is paying $55 million for antibodies that do not exist yet.

The molecules are real in silico. Earendil Labs designed them with protein models, scored them for binding and manufacturability, and handed Genentech a shortlist. In a wet lab, most of that shortlist is still a hypothesis. The rest of the money — up to roughly $1.5 billion in development, regulatory and sales milestones — appears only if the biology agrees.

What Genentech is actually buying

Strip the headline and the deal is narrow. Earendil runs antibody discovery and early research against a set of pre-agreed target combinations in oncology. Genentech, a member of the Roche Group, takes over global clinical development and commercialization. It pays tiered royalties on net sales.

Bispecific antibodies bind two targets at once. That lets them do what a standard antibody cannot. They can pull a T-cell onto a tumour cell, or block two escape routes in the same pathway. The format is now the busiest corner of oncology research, and it is hard to engineer. Two protein arms must fold, stay stable, and hit the right targets without falling apart on the manufacturing line.

That difficulty is the opening for a platform company. If a model can rank designs before a lab touches them, the cost per viable candidate falls.

The platform-as-product model

Earendil was founded in 2021 and is based in Wilmington, Delaware. Its pitch is a closed loop: predictive protein modelling, generative design, then high-throughput experimental validation that feeds results back into the model.

The commercial record is short and unusually dense. In January 2026, Sanofi committed up to $2.56 billion across autoimmune and inflammatory programmes, including up to $160 million in upfront and near-term payments. A 2025 licence with Sanofi had already covered two bispecific antibodies, HXN-1002 and HXN-1003. In February and March 2026, Earendil added manufacturing and payload-linker partnerships with WuXi XDC and WuXi Biologics. In March it raised $787 million from investors including DST Global, Dimension Capital and Sanofi, alongside a development fund backed by Hillhouse and Pfizer.

None of those programmes has produced an approved drug. The partners are buying access to a discovery engine, not a late-stage pipeline.

Genentech is a pioneer in defining modern biologics cancer medicine. Earendil Labs is building an AI-driven high-throughput biology platform that integrates AI directly into biologics research and development — from predictive protein modeling and generative protein design to rapid experimental validation.— Jian Peng, PhD, founder and CEO, Earendil Labs

Process over molecule.

What the milestone math hides

The headline says $1.5 billion. The wire transfer says $55 million. The gap is the entire story.

Milestone-heavy biotech deals price uncertainty into the structure. The upfront payment is what the buyer will spend before it learns anything new. Everything after that depends on data the programmes have not generated yet. So few preclinical programmes reach approval that the further a molecule sits from human trials, the more of the headline value is optionality rather than cash.

Company releases and trade coverage, 2026.

Compare the two Earendil pacts. Sanofi committed a larger headline number and a larger near-term cheque. Genentech committed less up front, but it took a narrower slice: oncology, where Earendil's own early clinical work stays in-house. The structure tells you what each partner valued. Sanofi bought breadth across autoimmune targets. Genentech bought one defined shot at a hard format.

The royalty line matters too. Tiered royalties on net sales are the part of the deal that survives every milestone reset. If a bispecific reaches the market, Earendil keeps a percentage of revenue without funding the late-stage trials.

Works in silico. In humans, the question stays open.

This is where the caution belongs. AI-designed molecules have reached the clinic. They have not, so far, produced a wave of approved drugs that validates the design step itself.

As we wrote in September, Enveda's $311 million Series E doubled its valuation as AI-discovered drugs reached the clinic. The same open question followed it: early human data is thin, and the endpoints that matter take years to read. The 2018 to 2021 wave of AI drug discovery companies is worth keeping in mind here — as cautionary comparison, not as a verdict on Earendil.

Earendil's position is different in one respect. It does not ask the market to believe the whole chain. It sells the discovery step to partners who carry the clinical risk. That lowers the burden of proof. It also caps the upside if one of the programmes becomes a blockbuster.

A single clinic-ready bispecific from the Genentech collaboration would be the first hard evidence. Until then, the deal is a signal about how pharma prices AI platforms, not proof that the platforms produce medicines.

The bigger pattern: pharma as the customer

Genentech did not sign this deal in isolation. In August 2026 it struck a $2.3 billion licensing deal with Hanmi Pharm for an obesity candidate and a global collaboration with DualityBio on next-generation antibody-drug conjugates. Roche's Genentech arm also announced a $750 million expansion of US manufacturing in August.

Large pharma behaves consistently here. Rather than build protein-design teams from scratch, it buys optionality from specialist platforms and keeps its own capital for clinical trials and manufacturing. Earendil sits on the sell side of that trade.

That position has a second potential buyer: public markets. Trade press reported in March that Earendil was weighing a Hong Kong listing. A company with two marquee collaborations and a $787 million balance sheet can tell an IPO story. Whether the story holds depends on data, not on the next signature.

What would change the thesis

Three events would move Earendil from platform story to pipeline story: a named clinical candidate from the collaboration, an undisclosed partner putting a molecule into Phase 1, and a repeat deal from a third top-20 pharma.

Does the $1.5 billion ever get paid?

Development scenarios