In 2026, Biohub became the central channel for Zuckerberg and Chan’s biomedical philanthropy, with AI, biological data, and new measurement tools at the center. For researchers, the shift matters because better cell measurements could help teams test ideas with more useful data. A virtual cell model is a computational model trained on biological information to predict how cells behave and respond to change.
The move marks a concentration of effort, not an announced end to other giving. CZI remains part of the work, while Biohub takes the lead in biomedical research. Priscilla Chan, CZI’s co-founder and co-CEO, leads its biomedical research and technology efforts. Mark Zuckerberg is also a CZI co-founder and co-CEO.
The central scientific question is whether richer measurements can make cell behavior predictable enough to guide better experiments.
Key Takeaways
- Biohub is the main channel for the couple’s biomedical work.
- AI models depend on better biological data and measurement tools.
- CZI and other philanthropic efforts continue.
- Model predictions need testing through experiments.
- Research targets are not promises of clinical benefit.
What Does the Zuckerberg Chan CZI Biohub AI Biomedical Shift 2026 Change?
The change is a sharper concentration of the couple’s giving around Biohub, not a declaration that CZI or every other philanthropic effort is ending. Biohub becomes the main channel for biomedical work, while CZI continues to provide infrastructure and support.
The couple launched CZI in 2015, and Biohub began work in 2016. Zuckerberg and Chan pledged 99% of their lifetime wealth toward their philanthropic efforts. Since 2016, they have given $4 billion to basic-science research, excluding operating costs for a large life-science computing cluster.
The giving shift puts most of the couple’s philanthropic resources behind Biohub, while other efforts continue.
That pledge and the $4 billion figure describe different things: one is a lifetime commitment, and the other is giving to date. The shift does not make Meta part of CZI or Biohub. Zuckerberg is Meta’s founder and CEO, while Priscilla Chan is CZI’s co-founder and co-CEO.
Biohub carries out its work through four legal and philanthropic entities:
- Chan Zuckerberg Biohub, Inc.
- Chan Zuckerberg Initiative, LLC
- Chan Zuckerberg Initiative Foundation
- Biohub Donor Advised Fund
In 2025, CZI announced plans to launch a Biohub in 2027 by combining CZ Biohub San Francisco and the CZ Institute for Advanced Biological Imaging. That consolidation remains a plan, not a completed merger. Keep the couple’s giving decision distinct from corporate commitments made by Meta as a company.
Who Is Funding the Virtual Biology Initiative, and What Does Each Partner Bring?
The $1.8 billion framework combines Biohub’s commitment with federal and commercial participation, while NIH coordinates data supported by earlier federal spending. The total includes prior federal investment, so it is not all new cash.
Announced in April 2026, the Virtual Biology Initiative (VBI) aims to coordinate open, AI-ready datasets and predictive models. Biohub’s October commitment totals $500 million. The other partners add funding, computing, research tools, or access to existing data.
| Contributor | Commitment or investment basis | Role |
|---|---|---|
| Biohub | $500 million | $400 million for measurement and engineering technology; $100 million for research outside Biohub |
| Department of Energy | More than $500 million over five years | Computing, lab tools, and national research facilities |
| NIH | More than $500 million in prior federal investment | Coordinates datasets, repositories, and knowledge bases |
| Google DeepMind, Isomorphic Labs, and Meta | $300 million collectively | Commercial investment in the initiative |
The Biohub announcement of the VBI commitments describes the combined effort. Federal investment supporting NIH’s data role came earlier, while DOE’s commitment covers five years.
These partners contribute in different ways, so the figures should not be read as four equal cash grants. NIH’s role centers on bringing existing data resources together for the initiative. Read the NIH row as a data-coordination role tied to prior investment, not as a new cash pledge.
How Will Virtual Biology Turn Cell Data Into Predictive Models?
A virtual cell model is meant to predict cellular behavior from biological measurements, then improve as predictions are tested against experiments. The model’s predictions must return to the lab for testing.
The models may combine genetic code, molecular information, images, and videos of activity within and between cells. Biohub is also working with NIH to standardize existing biomedical datasets for AI training. Standard formats can help researchers compare data gathered by different teams.
Spatial transcriptomics maps molecular activity in tissue
Spatial transcriptomics shows where molecular activity occurs inside intact tissue. That location matters because cells can behave differently beside other cell types than they do alone. Imaging and molecular measures together can give models a fuller view of those local patterns.
Cell-response screens measure the effects of change
Cell-response screens record how cells react to changes in their environment. The results let researchers compare a model’s forecast with what cells actually did. Those experiments can then help update later predictions.
Current cell datasets include hundreds of millions of cells. Biohub says an accurate predictive model may need datasets of billions, and eventually trillions, of cells. Its October 2026 targets call for a first VBI dataset in about a year and accurate predictive models within five years.
Commercially funded datasets may have embargo periods before public release. Government-funded work running in parallel has no such restrictions, though individual datasets may follow different release schedules. Treat the first-dataset and model-development dates as targets for a research program, not promised delivery dates for a treatment.
What New Measurement and Computing Tools Will the Initiative Build?
A model cannot learn the dynamics of living systems from a static snapshot alone. Biohub’s technology plans pair fine-detail imaging with tools that can change biology in controlled ways.
Biohub’s program includes cryo-electron tomography, which can reveal near-atomic detail inside cells. It also plans microscopy designed to image millions to billions of cells in living tissue. Those methods capture different scales, from structures inside a cell to changes across tissue.
Imaging cells from near-atomic detail to living tissue
Scott Fraser became president of the Chan Zuckerberg Imaging Institute in 2025. He also serves as president, dynamic imaging, and head of Biohub San Francisco. His roles connect imaging research with Biohub’s wider work, but the planned tools still need testing in research settings.
The Department of Energy’s Genesis Mission adds another set of capabilities. It will use exascale supercomputing, X-ray and neutron scattering, cryo-electron microscopy and tomography, and autonomous laboratories across the National Laboratory system. The Genesis Mission executive order sets out that federal plan.
Perturbing biology across scales
Biohub also plans tools to engineer and perturb biology, from molecular changes to whole-organism systems. A perturbation is a deliberate change that lets a team study what happens next. These tools can help researchers test cause and effect rather than only record what cells look like.
DOE’s computing and lab network can complement Biohub’s imaging and engineering work. Together, the capabilities could connect large-scale computation with experiments that measure real biological change. The practical test for these instruments is whether they capture changes at the scale the experiment needs to measure.
Which Four Scientific Challenges Shape CZI’s Next Decade?
CZI’s ten-year agenda has four connected challenges, spanning cell models, imaging, inflammation, and immune-based early intervention. Each challenge has a lead or planned leadership role, while the VBI remains a distinct data and modeling effort.
The four priorities set a wider science agenda for the decade:
- Build an AI-based virtual cell model, led by a head of AI.
- Develop imaging technologies, led by Scott Fraser.
- Measure inflammation in tissues in real time, led by Shana Kelley.
- Use the immune system for early disease detection, prevention, and treatment, led by Andrea Califano.
CZI said it was building a single-cell dataset of one billion cells to train models on cell behavior and gene function. That project supports the virtual-cell challenge, but it is not the same as the VBI’s broader coordination of data generation and predictive models.
CZI has said inflammation plays a role in more than 50% of human deaths each year. That figure explains the research focus; it does not show that planned tools have reduced disease or deaths. A head of data will also join the leadership team, alongside the planned head of AI.
For teams building AI systems, access to code and model design can shape how others test and adapt the work. The debate over open-weight AI models offers one example of how openness and control can affect research tools.
The leadership assignments show that the agenda combines AI, imaging, tissue measurement, and immune research rather than treating them as one project.
What Can Researchers Use Now, and What Early Results Show?
Researchers can encounter Biohub’s work through existing data projects and shared infrastructure, as well as through newer research outputs. These resources support discovery, but papers and models do not establish clinical benefit.
Open datasets and shared infrastructure
Biohub’s open-data projects include Tabula Sapiens, OpenCell, and Zebrahub. CELLxGENE and the CryoET Data Portal provide community data infrastructure. The resources serve different research needs, from cell data to images of cell structures.
| Resource | Resource type | Role in the ecosystem |
|---|---|---|
| Tabula Sapiens | Open-data project | Human cell data |
| OpenCell | Open-data project | Cell protein and imaging data |
| Zebrahub | Open-data project | Zebrafish development data |
| CELLxGENE | Community data infrastructure | Access to cell datasets |
| CryoET Data Portal | Community data infrastructure | Access to cryo-electron tomography data |
Research outputs and science ventures
CZI and Cold Spring Harbor Laboratory announced openRxiv in March 2025 as a permanent home for bioRxiv and medRxiv. CZI has supported more than 8,000 research publications since 2015. Selected 2026 outputs include “Genome-wide chromatin recording resolves dynamic cell state changes,” published in Cell Systems on September 2.
Other work includes “AI-guided CRISPR screening reveals therapeutic targets in psoriasis,” published in Nature Communications on July 13. BioRxiv preprints include “Language Modeling Materializes a World Model of Protein Biology” on June 4, “A multimodal perturbation atlas defines the phenotypic resolution of cellular morphology” on June 2, and “AI-Guided CRISPR Screen Accelerates Discovery of New Drug Targets” on February 28.
Biohub’s open discovery engine can generate, test, and refine billions of possible protein designs. Somite AI is building foundational AI models for cell therapy development, while Adaptyx Biosciences is developing continuous, multi-analyte molecular monitoring. Biohub AI engineering roles describe work on biological foundation models, reasoning, reinforcement learning, and multi-agent systems.
Research papers, preprints, and ventures show active work, not proven patient benefit.
Anthropic is pursuing biology with a wet lab, while the OpenAI Foundation launched a grant program exceeding $125 million for biological and medical datasets. Researchers who use AI coding assistants can compare tools in this JetBrains AI and Copilot overview, then review generated code and test changes. Treat papers and preprints as evidence of research activity, not as proof that a model has improved patient outcomes.
Conclusion
Biohub’s expanded role is a long-term bet on making biological systems more measurable and more predictable. Its significance depends on whether shared measurement and data systems produce models that help researchers design and test stronger experiments.
Research targets matter, but experimental checks will show whether the models make useful predictions. Future dataset releases offer a clear signal to watch, along with reports of experiments that test those predictions. Researchers can assess progress by tracking new data releases and asking whether model predictions hold up in experiments.
FAQ
How does the CZI challenge program differ from the Virtual Biology Initiative?
CZI’s challenge program sets four scientific priorities for the next ten years, while the VBI coordinates data generation and predictive models. The challenge program includes imaging, inflammation, and immune research beyond cell modeling. The VBI focuses on building and organizing data that can train models. Its planned dataset and model dates are separate research targets, not deadlines for all four CZI challenges.
What makes a virtual cell model different from a cell atlas?
A virtual cell model aims to predict cell behavior and responses to change, while a cell atlas organizes observations about cells. An atlas can show which cell types appear in a tissue or which genes they express. A model uses such information to make predictions that researchers can test. Predictive ability does not mean the model is ready for clinical use.
Is Biohub’s work limited to human cells?
No, Biohub’s work extends beyond human-cell prediction. Its projects include protein design, and its planned engineering tools span molecular changes through whole-organism systems. Existing data resources also include Zebrahub, which focuses on zebrafish development. These applications serve different research questions, so they should not be treated as one human-cell model or one clinical program.
What does “AI-ready” mean for biomedical data?
AI-ready biomedical data is organized in a standard way that lets teams combine and compare measurements. That can include molecular data, images, and records of how cells respond to experimental changes. Standardizing datasets helps models work across information gathered by different groups. It does not mean every dataset is complete, public, or released on the same schedule.
How do Biohub and Meta relate to the philanthropic shift?
Biohub is the central channel for Zuckerberg and Chan’s biomedical philanthropy, while CZI continues to provide support and infrastructure. Meta is a separate company led by Zuckerberg, not the philanthropic organization. Meta is also one of three commercial participants, with Google DeepMind and Isomorphic Labs, investing collectively in the VBI. Those roles belong to different organizations and should not be merged.
