Building open AI to cure or prevent all disease by 2110 (www.latent.space)

The Chan Zuckerberg Initiative (CZI) announced a major push to marry massive compute, new models and wet-lab imaging to build a “Virtual Cell” and a “Virtual Immune System” as part of a long‑horizon plan to prevent or cure all disease by 2110. CZI’s Biohub — funded at roughly $1B/year — has acquired EvoScale (the team behind ESM‑3), deployed a 10,000‑GPU cluster for bio research, open‑sourced a large human cell atlas, and released new AI tools including VariantFormer (translates personal variants into tissue‑specific gene activity), CryoLens (pretrained cryo‑ET model for unsupervised structural similarity), and scLDM (high‑fidelity single‑cell data generator), alongside GREmLN and rBio. They’re also investing in custom hardware (CryoET microscopes) and colocated interdisciplinary labs to close the loop from atoms→images→bits→models. For the AI/ML community this is significant because it shifts biology from “parts lists” (genomes, AlphaFold structures) to system‑level, predictive models — enabled by scale (10k GPUs), large multimodal datasets, and pretrained domain models — promising to make in‑silico experiments orders of magnitude faster and cheaper. Technical implications include new opportunities for large‑scale pretraining, multimodal integration (imaging + single‑cell + sequence), generative single‑cell simulation, and model‑led experimental design that can accelerate therapeutic discovery (e.g., immune engineering, CAR‑T, cell‑based diagnostics). If successful, this frontier lab model could become the blueprint for how AI and biology converge to tackle complex, system‑level biomedical problems.