The EvolutionaryScale team joins Biohub

Background

Biohub has existed since 2016, and describes itself as having built the largest single-cell datasets used in biological research ^[1]. Single-cell data records what individual cells are doing rather than averaging a tissue into one reading, which is the kind of raw material that machine-learning models need if they are to learn patterns of cell behaviour.

On 6 November 2025, Mark Zuckerberg and Priscilla Chan announced a new Biohub initiative combining frontier AI with frontier biology ^[1]. The phrase points at two things being pushed forward at once: the largest and most capable AI systems, and biological research at its most ambitious scale.

What happened

As part of the initiative, the team at EvolutionaryScale joined Biohub ^[1]. Alex Rives, EvolutionaryScale's co-founder and chief scientist, became Biohub's head of science ^[1]. The arrangement brings a group built around AI research inside an organisation whose standing comes from large biological datasets, and places the incoming group's scientific lead in charge of the science agenda.

Key details

Biohub said it would expand its compute capacity tenfold, to 10,000 GPUs by 2028 ^[1]. Compute is what AI models are trained on, so the number sets a scale for the work the organisation intends to do rather than describing a finished system.

Biohub also committed to four grand challenges: a unified AI model of the cell, advanced imaging, inflammation instrumentation and immune-system reprogramming ^[1]. Each names a target rather than a single experiment — a model that represents a whole cell rather than one component, better ways of imaging biological tissue, instruments aimed at inflammation, and attempts to reprogram the immune system.

The announcement additionally launched a Virtual Immune System effort and released three models, named VariantFormer, CryoLens and scLDM ^[1]. Issuing models alongside the announcement means outside researchers can start using them without waiting for a separate release.

Why it matters

Zuckerberg said the goal was to help scientists cure or prevent all diseases this century, and that AI made that possible much sooner ^[1]. That framing is broader than any single disease or any single model: it treats disease in general as the target, with AI as the thing that changes the timescale.

The announcement is unusual in putting several kinds of commitment in one place. There is an organisational move — a research team and its scientific lead arriving ^[1] — alongside an infrastructure number, a list of long-range scientific goals and a set of released models ^[1]. Read together, they describe a programme rather than one paper or one product.

The compute target is also the part with a date attached: ten times the current capacity, reaching 10,000 GPUs by 2028 ^[1]. Everything else in the announcement is a direction of travel, so the expansion is the concrete commitment against which progress can be checked.

The grand challenges are similarly structured as long-running goals rather than deliverables ^[1]. A unified model of the cell, in particular, is the kind of target that would take many datasets and many groups to reach; a single organisation can set the goal, but not finish it alone.

The models released with the announcement are the smallest and most immediately usable piece ^[1]. They are also the part that other researchers can test, compare and build on, which is how a research direction becomes a field rather than a press release.