Mark Zuckerberg & Priscilla Chan: How AI Will Help Cure Disease

What It Covers

Mark Zuckerberg and Priscilla Chan outline how the Chan Zuckerberg Initiative's Biohub network is combining frontier AI with frontier biology to build shared scientific tools — including cell atlases, virtual cell models, and protein models — targeting disease elimination by end of century, now potentially sooner.

Key Questions Answered

• •Tool-first science strategy: Rather than funding individual disease therapies, CZI invests $100M–$1B over 10–15 year horizons to build shared infrastructure — datasets, imaging systems, AI models — that the entire scientific community can access freely. This approach mirrors how the microscope and telescope unlocked prior scientific revolutions by enabling observation of previously invisible phenomena.
• •CellxGene network effect: CZI built CellxGene originally as a single-cell data annotation tool to solve a workflow bottleneck. Standardized formatting caused organic adoption across the broader research community, resulting in a cell atlas where 75% of contributed data came from external researchers — not CZI-funded labs — demonstrating how open tooling creates compounding scientific returns.
• •Virtual cell model hierarchy: CZI is building virtual cell models in layers — proteins first (via Evolutionary Scale partnership), then cellular behavior, then systems like a virtual immune system. Models like Variantformer predict CRISPR edit outcomes; a diffusion model generates synthetic rare cell configurations. This hierarchy lets researchers test hypotheses computationally before expensive wet lab experiments.
• •Biohub geographic specialization: Three Biohubs address distinct biological challenges: San Francisco focuses on deep imaging and transcriptomics, Chicago studies cell communication within tissues and inflammation, New York works on cell engineering for in-body signal detection. Each hub partners with local universities — UCSF, Stanford, Berkeley, University of Chicago — to enable cross-disciplinary collaboration between biologists and engineers.
• •Compute as the new lab space: CZI operates a 1,000-GPU compute cluster with plans to scale to 10,000 GPUs, which it opens to external scientists via a competitive application process. Individual academic labs typically operate with tens of GPUs, making this cluster access a meaningful capability multiplier for researchers pursuing questions that require large-scale computational resources unavailable through standard grant funding.