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

Summary

• The Biohub thesis is that shared scientific tools—not another round of small grants—are the highest-leverage route to faster cures. Mark Zuckerberg said major breakthroughs usually follow new ways to observe phenomena, yet tools such as imaging systems and virtual-cell models can require $100 million to $1 billion over 10–15 years. “We’re not going to cure all diseases”; the strategy is to help the scientific community do it.
• AI changes the timetable only if frontier models and frontier biology form a closed data flywheel. Biohub intends to design experiments and instruments around model blind spots, generate purpose-built data, retrain the models, and repeat. That bridges a striking cultural split: biologists considered curing disease “crazy ambitious,” while AI researchers thought it was “kind of boring—that’s just automatically going to happen.”
• CELLxGENE demonstrates how open infrastructure can acquire a network effect larger than its original funder. Built to clear an annotation bottleneck, it standardized formats and metadata across single-cell labs; CZI funded only 25% of the resulting resource, while the broader community contributed 75%. Vineeta Agarwala’s shorthand: “Come for the annotation, stay for the virtual cell model.”
• Virtual cells could expand biotechnology’s risk budget before expensive wet-lab work begins. Vineeta Agarwala argued that even directional predictions could let researchers test bolder hypotheses in silico before costly experiments, without risking years of work, publication, or tenure. Zuckerberg likened the model to “the new fruit fly” and said a perfect simulation is unnecessary; a directional signal could still be useful. The ambition is human-relevant modeling, tempered by the maxim: “All models are wrong. Some are useful.”
• Precision medicine here means treating common diseases as collections of individually rare biology. Chan’s framing was categorical: “Most diseases should be thought of as rare diseases,” because today’s hypertension and depression treatments still rely heavily on trial and error. Connecting a mutation to downstream cells, protein expression, drug targets, and predicted off-target effects could enable more precise diagnostics and therapies.
• CZI is concentrating its philanthropy around a unified operating Biohub that combines AI, data generation, and biological research. EvolutionaryScale researchers who formerly worked at Meta on protein-folding models are joining a Biohub, with its leader set to run the broader science program. Chan said, “The Biohub is really going to be the main thrust of our philanthropy,” while education and local-community work continue. They did not present centralization as a model for all science; decentralized work and outside labs remain important.
• The next laboratory expansion is computational: Biohub plans to move from roughly 1,000 GPUs toward the 10,000 range. Outside scientists can apply to use that capacity for questions individual labs—with only tens of GPUs—cannot tackle. As Zuckerberg put the allocation constraint, “The GPUs are somewhat zero-sum. The data isn’t.”

Deep dive

1. The cure-all mission became an infrastructure strategy

• Chan traced the mission to pediatric training at UCSF, where families might receive a gene name or a generic PDF but little actionable guidance. Basic science was the missing “pipeline of hope”: without understanding disease mechanisms, clinical care could not advance beyond translating a few uncertain facts.

• Zuckerberg clarified the division of labor: “We’re not going to cure all diseases”; Biohub aims to help scientists cure them. His historical premise is that major breakthroughs are generally preceded by tools such as microscopes and telescopes—otherwise researchers are “coding without being able to step through the code and debug things.”

• Conventional funding could not plausibly deliver the century goal by awarding every lab its next-best grant. Shared tools, standardized datasets, imaging systems, and virtual-cell models require coordinated bets of roughly $100 million to $1 billion over 10–15 years—an underfunded space between near-term NIH projects and century-scale ambition.

2. Grand challenges translate ambition into investable time horizons

• Chan’s selection rule for a Biohub challenge is demanding but bounded: “I see a path,” a credible leader exists, and enough ambiguity remains to justify risk. A fully solved problem is insufficiently ambitious; a successful 10–15-year program should produce returns greater than initially expected.

• The three sites encode distinct capabilities: New York engineers cells to detect, record, or act on signals; Chicago studies tissue construction and cell-to-cell communication, including inflammation; San Francisco develops deep imaging and transcriptomics. Partner universities supply complementary labs while Biohub teams work collaboratively across traditional boundaries.

• Success is enabling an “explosion” of precision-medicine builders, not manufacturing every therapy internally. Chan described the chain from a genetic variant—including the dreaded “variant of unknown significance”—through downstream cell effects and protein expression to a target, while also predicting where a drug might interact elsewhere in the body.

3. CELLxGENE turned an annotation fix into shared infrastructure

• Zuckerberg called it “crazy” that biology in 2025 still lacks a “periodic table of elements equivalent.” The single-cell atlas began a decade earlier with methodology funding and seeded datasets, before virtual-cell models had become the explicit destination.

• CELLxGENE itself emerged from a mundane workflow failure: researchers could generate single-cell data faster than they could annotate it. Biohub built an annotation interface, published the collected data, and unintentionally standardized formats and metadata because participating labs adopted the same tool.

• The resulting feedback loop exceeded CZI’s own contribution. Mark Zuckerberg said CZI had funded 25% of the shared resource and the wider community supplied 75%, contributing millions of cells because sharing had become useful and operationally easy.

• Vineeta Agarwala supplied the downstream example: a portfolio company studying idiopathic pulmonary fibrosis queries millions of healthy and diseased cells through CELLxGENE atlases, identifies fibroblasts, and examines their gene expression for possible drug targets. The example shows open tooling reaching startups, pharma, and R&D communities, whose users may not necessarily publish papers about or directly credit the tools.

4. Virtual cells lower the cost of asking risky biological questions

• The proposed architecture climbs from proteins through intracellular structures and whole cells toward systems such as a virtual immune system. Biohub sites generate novel datasets, individual models learn distinct capabilities, and those models are progressively combined into a more general simulator for hypothesis generation and drug discovery.

• Zuckerberg described several early capabilities: Variantformer learns from cells edited with CRISPR and predicts resulting states; a diffusion model generates a synthetic version of a requested cell type; and a cryo-EM model adds spatial understanding. The objective is not one monolithic model but increasingly integrated views of rare cellular configurations.

• Vineeta Agarwala’s case is economic as much as scientific: wet-lab experiments consume money and time, so researchers must favor projects likely to preserve funding, publications, and careers. High-quality simulation could test riskier ideas in silico first. On fidelity, Zuckerberg said the model need not be 100% accurate; even a directional signal could usefully derisk the next experiment.

• A biological reasoning model would need to explain how and why states evolve, not merely return correlations, though Zuckerberg stressed that this work is “quite early.” His hierarchy matters: cellular predictions remain shallow without protein-level understanding, while protein and cell models together can support more complex simulations.

5. Unifying the teams closes the model-to-experiment loop

• Biohub’s reorganization brings previously decentralized biological research, software, datasets, and AI under one operating team. Chan’s mechanism is a “complete flywheel”: when a model exposes gaps or blind spots, scientists can design the next dataset and feed its unusually rich metadata directly back into training.

• The hosts noted that domain-specific models have proved unexpectedly strong even where data is open; knowing the exact problem and user still matters. Biohub extends that lesson to interfaces: CELLxGENE deliberately lowers computational barriers, and future virtual models should let immunologists, for example, bring their expertise into neurodegeneration research.

• Zuckerberg cautioned that this is not a claim that all science should be centralized. The Biohub model is intended to fill a space where collaboration and shared resources can unlock value, while decentralized work and outside labs remain necessary.

6. Compute and measurable use now anchor the next decade

• Biohub may add sites and a central AI team, but square footage is no longer the main unit of expansion: “We’re expanding our compute.” Individual academic labs may hold tens of GPUs; shared clusters at 1,000 and eventually 10,000 permit qualitatively different questions and seed collaborations with outside scientists.

• Chan admitted spending her early philanthropic years envying the market’s clear feedback. Ten years in, she sees a usable signal: Biohub achieved what it promised, and its projects often delivered more than expected. The operating principle is to maintain “a long time horizon, but be impatient” through rapid iteration amid ambiguity.

• Zuckerberg said advances in AI should make the end-of-century disease mission achievable “significantly sooner,” prompting the continued concentration on science. The closing discussion placed Biohub in the infrastructure layer between basic research, biotech therapy creation, scaled pharma delivery, and public-health access. Ben Horowitz’s test was: “If we didn’t exist, would it be a problem? Yes.”