The Chan Zuckerberg Initiative is now focusing on using AI to speed up biology research, with a huge $10 billion commitment. They are building giant datasets and powerful AI tools to understand the cells in our bodies faster, hoping to solve problems like fighting disease and measuring inflammation in real time. Their labs, called Biohubs, are filled with computers and top scientists working together to build a virtual cell and make discoveries much quicker than before. They share their data and tools openly with the world, aiming to help researchers everywhere solve medical mysteries soon.

CZI shifts focus to AI-powered biology, pledges $10 billion

Chan Zuckerberg Initiative Restructures to Focus on AI and Scientific Research with Biohub is no longer a headline but the day-to-day reality inside the philanthropy's labs. In late 2025, Mark Zuckerberg and Priscilla Chan redirected billions toward AI-powered biology, betting that computation can compress years of wet-lab work into days.

The new strategy is simple: build ever larger biological datasets, train frontier models on them, and share the resulting tools with universities worldwide. Within months, the Biohub network in San Francisco, New York, and Chicago began hiring protein language model experts and purchasing GPUs at a scale most universities cannot match.

The organization has already secured a decade-long research budget of 10 billion dollars, more than double its previous pledge. According to the official CZI Science Overview, 10,000 GPUs will be online by 2028 to support model training and open API access.

Grand challenges that guide the pivot

CZI leaders distilled their ambition into four challenges that they believe cover the cellular universe:

• Harnessing the immune system
• Measuring inflammation in real time
• Building a virtual cell with AI
• Deciphering cellular networks

Each challenge is assigned to cross-disciplinary teams spanning computer science, synthetic biology, and clinical medicine. The New York Biohub, for example, recently launched projects in immune cell engineering and single-cell analytics in partnership with Columbia, Rockefeller, and Yale, as confirmed by the CZ Biohub New York announcement.

Chan Zuckerberg Initiative Restructures to Focus on AI and Scientific Research with Biohub - what changes on the ground

Scientists working inside the hubs now start with computation, not pipettes. Engineers feed terabytes of single-cell RNA data into models such as GREmLN, which highlights cancer gene interactions earlier than standard tools. Imaging specialists then validate the predictions in living tissues, tightening a rapid feedback loop.

TranscriptFormer and Cytoland further shrink analysis time. These AI tools sift through tissue images and sequencing reads in minutes, pointing researchers toward unusual patterns linked to neurodegeneration or aggressive tumors.

Infrastructure and collaboration at unprecedented scale

The hardware build-out matters. GPU clusters, managed like hyperscale cloud farms, enable experiments that demand billions of parameters. CZI is also generating what it calls the largest map of human cell types ever assembled. The data flow continually refines the virtual cell model, a long-term project that aims to simulate intracellular processes the way climate models simulate weather.

Collaboration extends outside CZI's walls. Bridge2AI at NIH and similar programs worldwide seek interoperable datasets; Biohub investigators coordinate standards so models trained in Palo Alto remain useful in Chicago or Abu Dhabi. Private pharma firms, watching the 30 percent annual growth in AI drug discovery spending, queue for partnership slots.

Early evidence of impact

Researchers at the Chicago hub have already demonstrated real-time inflammation sensors that detect cytokine surges in mouse tissue within seconds. Preliminary results, shared at internal reviews, suggest the method could forecast autoimmune flare-ups days before symptoms appear. Meanwhile, the virtual cell prototype predicts T cell exhaustion markers with an accuracy that surprised outside reviewers from Stanford and MIT, according to reporting by Pulse2.

None of these tools remain locked behind paywalls. APIs, documentation, and benchmark datasets go live on CELLxGENE and GitHub as soon as peer review clears. CZI leadership argues that open access accelerates replication and avoids siloed discovery.

The next milestones are clear. By 2026, Biohub teams plan to release a second generation of models trained on one billion single-cell profiles. By 2027, they aim to integrate pathology images and proteomics, edging closer to a full-fidelity virtual cell. If the timeline holds, disease modeling could soon move from lab benches to GPU clusters, changing how biologists everywhere design experiments.

Chan Zuckerberg Initiative Restructures to Focus on AI and Scientific Research with Biohub is no longer a headline but the day-to-day reality inside the philanthropy's labs. In late 2025, Mark Zuckerberg and Priscilla Chan redirected billions toward AI-powered biology, betting that computation can compress years of wet-lab work into days.

The new strategy is simple: build ever larger biological datasets, train frontier models on them, and share the resulting tools with universities worldwide. Within months, the Biohub network in San Francisco, New York, and Chicago began hiring protein language model experts and purchasing GPUs at a scale most universities cannot match.

The organization has already secured a decade-long research budget of 10 billion dollars, more than double its previous pledge. According to the official CZI Science Overview, 10,000 GPUs will be online by 2028 to support model training and open API access.

Grand challenges that guide the pivot

CZI leaders distilled their ambition into four challenges that they believe cover the cellular universe:

• Harnessing the immune system
• Measuring inflammation in real time
• Building a virtual cell with AI
• Deciphering cellular networks

Each challenge is assigned to cross-disciplinary teams spanning computer science, synthetic biology, and clinical medicine. The New York Biohub, for example, recently launched projects in immune cell engineering and single-cell analytics in partnership with Columbia, Rockefeller, and Yale, as confirmed by the CZ Biohub New York announcement.

Chan Zuckerberg Initiative Restructures to Focus on AI and Scientific Research with Biohub - what changes on the ground

Scientists working inside the hubs now start with computation, not pipettes. Engineers feed terabytes of single-cell RNA data into models such as GREmLN, which highlights cancer gene interactions earlier than standard tools. Imaging specialists then validate the predictions in living tissues, tightening a rapid feedback loop.

TranscriptFormer and Cytoland further shrink analysis time. These AI tools sift through tissue images and sequencing reads in minutes, pointing researchers toward unusual patterns linked to neurodegeneration or aggressive tumors.

Infrastructure and collaboration at unprecedented scale

The hardware build-out matters. GPU clusters, managed like hyperscale cloud farms, enable experiments that demand billions of parameters. CZI is also generating what it calls the largest map of human cell types ever assembled. The data flow continually refines the virtual cell model, a long-term project that aims to simulate intracellular processes the way climate models simulate weather.

Collaboration extends outside CZI's walls. Bridge2AI at NIH and similar programs worldwide seek interoperable datasets; Biohub investigators coordinate standards so models trained in Palo Alto remain useful in Chicago or Abu Dhabi. Private pharma firms, watching the 30 percent annual growth in AI drug discovery spending, queue for partnership slots.

Early evidence of impact

Researchers at the Chicago hub have already demonstrated real-time inflammation sensors that detect cytokine surges in mouse tissue within seconds. Preliminary results, shared at internal reviews, suggest the method could forecast autoimmune flare-ups days before symptoms appear. Meanwhile, the virtual cell prototype predicts T cell exhaustion markers with an accuracy that surprised outside reviewers from Stanford and MIT, according to reporting by Pulse2.

None of these tools remain locked behind paywalls. APIs, documentation, and benchmark datasets go live on CELLxGENE and GitHub as soon as peer review clears. CZI leadership argues that open access accelerates replication and avoids siloed discovery.

The next milestones are clear. By 2026, Biohub teams plan to release a second generation of models trained on one billion single-cell profiles. By 2027, they aim to integrate pathology images and proteomics, edging closer to a full-fidelity virtual cell. If the timeline holds, disease modeling could soon move from lab benches to GPU clusters, changing how biologists everywhere design experiments.