How they use Nebius AI Cloud

Google DeepMind participated in CRISPR-GPT, a joint research effort with Stanford and Princeton that used Nebius compute. The project developed an LLM agent system to assist scientists with gene-editing experiment design and analysis. Nebius describes supporting rapid model screening and fine-tuning with a single node containing eight NVIDIA H100 GPUs. In a November 2025 podcast, Nebius’s Ilya Burkov said the joint effort used its compute for training and running models.

The research paper identifies Google DeepMind’s Denny Zhou as a contributor to model construction, deployment and evaluation. It acknowledges supercomputing resources supplied by Nebius Group N.V. CRISPR-GPT combines scientific knowledge, retrieval and external tools to help researchers select CRISPR systems, design guide RNAs, plan experiments and analyse results. TD SYNNEX states that scientists from Stanford, Princeton and Google DeepMind automated gene-editing experiments on Nebius. It reports a 288-case evaluation framework and 80–90% first-attempt efficiency for novice researchers.

Burkov describes a research workflow combining computational work with wet-lab testing, covering gene knockout, base editing and prime editing. The interactive agent reasons with researchers about methods and gene sets.