In 2016, Priscilla Chan and Mark Zuckerberg announced they would dedicate 99% of their Facebook shares — then worth around $45 billion — to the Chan Zuckerberg Initiative, with one of its central stated goals being to “cure, prevent, or manage all disease by the end of the century.” That felt audacious. Then they compressed the timeline. The current CZI framing targets having the tools to defeat most diseases within a single generation — roughly by 2030 for the foundational science layer. Most people dismissed this as billionaire optimism. They were probably wrong to dismiss it.
Here’s what changed: AI actually started working in biology. Not as a metaphor, not as a slideshow feature — as a tool that can predict protein structures, design novel molecules, read pathology slides, and now help orchestrate multi-step scientific reasoning in ways that would have been pure science fiction in 2015. The convergence of foundation models, biological datasets at unprecedented scale, and serious institutional capital is producing real outputs. CZI’s bet is that if you fund the infrastructure layer — compute, open datasets, AI tools for researchers — you can compress decades of biological discovery into years.
Whether 2030 is the right date is almost beside the point. The question worth asking is: what, specifically, does AI make possible in medicine and biology right now, and how does CZI’s model fit into that picture?
What CZI Is Actually Building
CZI isn’t a pharma company. It’s not running clinical trials or selling drugs. What it does is fund and build the infrastructure that scientific research runs on — and increasingly, that infrastructure is AI-native.
The flagship project is the Chan Zuckerberg Biohub, a nonprofit research institute with hubs in San Francisco, Chicago, and New York. But the more technically significant investment is CZ CELLxGENE, an open platform hosting one of the largest collections of single-cell RNA sequencing data in the world — over 50 million cells as of early 2026. Single-cell data is the raw material for understanding how individual cells behave in health and disease. That dataset is the kind of thing that makes large biological foundation models possible.
CZI also co-funds OpenCell, a proteomics atlas project, and has invested heavily in tools like scVI (single-cell Variational Inference), a deep learning framework for analyzing single-cell data that’s now used by research groups globally. These aren’t headline-grabbing announcements — they’re the unglamorous infrastructure bets that tend to matter most over time.
The model CZI is running is essentially: build open platforms, fund open science, attract top researchers, and let AI amplify their output. The goal isn’t to own the discovery. It’s to lower the cost and time of discovery across the entire field.
Where AI in Biology Is Actually Working Right Now
It’s worth separating hype from function here, because both exist in abundance.
Protein structure prediction is the clearest success story. DeepMind’s AlphaFold2 — and now AlphaFold3 — can predict the 3D structure of proteins with accuracy that took experimental methods years to achieve. The AlphaFold Protein Structure Database now covers over 200 million protein structures. This is real. Researchers at the Broad Institute, the Wellcome Sanger Institute, and hundreds of university labs are using it daily. Demis Hassabis has said this is the result he’s most proud of from DeepMind’s entire history, and the 2024 Nobel Prize in Chemistry awarded to AlphaFold’s development team validated that assessment.
Drug discovery acceleration is real but messier. Insilico Medicine used AI to identify a novel drug candidate for idiopathic pulmonary fibrosis — INS018_055 — that reached Phase 2 clinical trials faster than most traditional pipelines. Recursion Pharmaceuticals is running high-throughput biological experiments and using ML to identify patterns across millions of cellular images. These are real companies with real pipelines. But “AI-discovered drug in clinical trials” is different from “AI-discovered drug that works and gets approved.” The attrition rate in clinical development remains brutal. AI is compressing the pre-clinical phase; it hasn’t yet cracked the clinical trial bottleneck.
Pathology and imaging is arguably the most mature clinical AI application. Google’s LYNA (Lymph Node Assistant) detects metastatic breast cancer in pathology slides with accuracy matching or exceeding experienced pathologists in controlled settings. Paige.AI received the first FDA clearance for an AI-based pathology tool. Radiology AI from companies like Aidoc and Viz.ai is actively deployed in hospital systems detecting strokes, pulmonary embolisms, and intracranial hemorrhages. This is in production, not in a lab.
Genomics and multi-omics integration is where things get complicated and exciting simultaneously. Models like Evo (from Arc Institute, partially CZI-backed) are training on DNA sequences the way LLMs train on text — learning the “grammar” of genomes. Early results suggest these models can predict functional effects of mutations and even design novel biological sequences. It’s early. But the trajectory is clear.
The Infrastructure Bet: Why Open Science Matters More Than Any Single Breakthrough
One of the most underrated aspects of CZI’s strategy is the open science commitment. In an era where AI companies are racing to lock up data and models, CZI is explicitly funding open datasets, open-source tools, and open publication. This is philosophically and practically significant.
Peter Diamandis and Steven Kotler, in The Future Is Faster Than You Think, argue that one of the key accelerants of exponential change is the democratization of tools — making what was once available only to the elite available to anyone. CZI’s model is a direct instantiation of that principle applied to biological research. When a researcher in Lagos or Bangalore can access the same single-cell datasets and AI tools as someone at Harvard Medical School, you’ve fundamentally changed the geometry of who can contribute to discovery.
This isn’t charity. It’s a leverage strategy. The more researchers using your platform, the more signal you get about what works, the more the tools improve, and the faster the field moves. Open science at scale is a compounding bet.
The contrast with Big Pharma’s historical model — patents, proprietary data, closed pipelines — is stark. Neither model is purely right. Pharmaceutical companies need revenue incentives to fund the expensive, risky business of clinical development. But the pre-competitive research layer — understanding basic disease mechanisms, building reference atlases, developing shared tools — benefits from openness. CZI is essentially subsidizing that layer for the entire field.
What the Timeline Actually Means
When CZI says “defeat most disease by 2030,” the technically honest interpretation is: by 2030, have the scientific tools and understanding to make most diseases manageable or curable, even if the actual treatments take longer to reach patients. That’s a different claim than “no one will get cancer in 2030.”
Andrej Karpathy has talked about the concept of “software 2.0” — where instead of writing explicit rules, you train neural networks on data and the system learns the rules itself. Biology is experiencing its own Software 2.0 moment. The old model was: hypothesis → experiment → publi