Accelerating and Scaling Biological Sciences via AI RFA (Discontinued)

Chan Zuckerberg Initiative Accelerating and Scaling Biological Sciences with AI RFA (GPU Cluster Allocation) is sponsored by Chan Zuckerberg Initiative (CZI). CZI's Accelerating and Scaling Biological Sciences with AI RFA provides allocations of high-performance GPU compute to researchers building large-scale AI/ML models for the biological sciences that cannot be created with conventional university resources.

The Chan Zuckerberg Initiative (CZI) invites proposals to build large-scale AI/ML models for biology that cannot be created with conventional university computing resources. The RFA provides an in-kind allocation of CZI's GPU cluster (minimum request of 96 GPUs) rather than cash, supporting the training of large biological foundation models and other compute-intensive AI-for-science research. The program is part of CZI's expanded AI-powered biomedical science strategy aimed at helping cure, prevent, or manage all diseases.

CZI's Accelerating and Scaling Biological Sciences with AI RFA provides allocations of high-performance GPU compute to researchers building large-scale AI/ML models for the biological sciences that cannot be created with conventional university resources. Awardees receive an in-kind allocation (minimum 96 GPUs) on CZI's CoreWeave-hosted cluster of 1,024 Nvidia H100 GPUs in a DGX SuperPod configuration with VAST fast data storage and a fully managed Kubernetes environment, for projects up to one year. No cash funds are provided. Priority goes to projects aligned with CZI's virtual cell work, though all proposals advancing CZI's mission to cure, prevent, or manage all diseases are considered. Awards require open-science and code-sharing commitments (permissive licenses), de-identified data only, and rolling review.

CZI's Accelerating and Scaling Biological Sciences with AI RFA awards compute, not money, and that distinction shapes everything about who should apply. Successful projects receive an allocation on CZI's cluster of 1,024 NVIDIA H100 GPUs in an NVIDIA DGX SuperPOD configuration; the RFA is explicit that there are no cash funds, financial contributions or fees of any kind. Salaries, data acquisition and everything else must already be funded. The floor is a real filter: projects must use a minimum of 96 GPUs during their large training phase, which excludes ordinary fine-tuning work and targets teams genuinely training large-scale AI and machine learning models for the biological sciences. Priority goes to virtual cell modelling, consistent with CZI's mission to cure, prevent or manage all diseases by the end of the century. Allocations run for a maximum of one year from project start. Two conditions deserve attention before applying. First, open science is mandatory - code and model weights must be shared - so teams with commercial or embargo constraints should not apply. Second, only de-identified data may be used, and institutional sign-off is required. Eligibility is restricted to U.S.-based domestic nonprofit organisations including colleges, universities, hospitals, research labs and government agencies; for-profit organisations are ineligible, PIs and Co-PIs must hold faculty or equivalent independent investigator status at a U.S. nonprofit institution, and Meta employees are ineligible. Review is on a rolling basis against periodic review dates, with the practical warning that the cluster may reach full allocation before the final deadline of a cycle - applying early in a cycle materially improves the odds. Applicants should confirm current cycle dates with CZI at sciencegrants@chanzuckerberg.com.

CZI's Accelerating and Scaling Biological Sciences with AI request for applications awards GPU compute rather than money, allocating time on a cluster of 1,024 Nvidia H100 GPUs to academic teams building large-scale AI and machine learning models for biology. The minimum request is 96 GPUs, which sets a floor on the ambition of eligible projects: this is aimed at training substantial models, not fine-tuning. CZI prioritises work aligned with building virtual cells, meaning multi-scale neural network models that simulate molecular and cellular behaviour, though any biomedical research supporting CZI's disease-prevention mission qualifies. Projects must complete within a maximum one-year timeline and demonstrate they can genuinely scale to 96 or more GPUs, a technical bar that eliminates many otherwise strong proposals. The open-science conditions are real obligations rather than boilerplate: grantees must release code under MIT, BSD or Apache licences, make datasets publicly available, follow preprint publication protocols, and file technical reports. Applications were accepted on a rolling basis with review dates on the fifteenth of January, February, March and April, and CZI reserved the right to close the portal early once cluster capacity was fully allocated - which is the practical constraint, since the resource is finite and allocations are not renewed automatically. Teams should check the CZI RFA page for the current cycle.