NVIDIA launches genomic NIM

DNA is the cell’s instruction code, and NVIDIA has turned its Evo 2 genome model into a deployable service on Amazon SageMaker JumpStart. (aws.amazon.com) Amazon Web Services said on February 24, 2026 that the NVIDIA Evo-2 NIM microservice is now listed in SageMaker JumpStart, which lets customers launch prebuilt models from Amazon’s managed machine-learning platform. AWS said the service is aimed at drug discovery workflows and can be deployed through SageMaker Studio. (aws.amazon.com) NVIDIA describes NIM, short for NVIDIA Inference Microservices, as packaged containers for running models in production. AWS added SageMaker support for NVIDIA NIM in March 2024, first for large language models, and is now extending that setup to bioscience models including Evo 2. (aws.amazon.com) A genomic foundation model works like an autocomplete system for biological code: it learns patterns in long strings of DNA, RNA, and protein sequences, then predicts what changes might do or generates new candidate sequences. Arc Institute says Evo 2 can model biology from single-letter nucleotide changes up to whole genomes. (arcinstitute.org) Arc Institute and NVIDIA introduced Evo 2 on February 19, 2025 as a model trained on more than 9.3 trillion nucleotides from more than 128,000 genomes across bacteria, archaea, and eukaryotes. Arc said the model can identify disease-causing mutations in human genes and design genome-length sequences for simple organisms. (arcinstitute.org) NVIDIA’s BioNeMo documentation lists Evo 2 at 40 billion parameters and says it is ready for commercial use. The company says the model can handle tasks from mutation-effect prediction to generating coding-rich sequences up to 1 million tokens in length. (docs.nvidia.com) The new piece is not the model itself but the runtime around it. AWS said the SageMaker version can be chained with other NVIDIA microservices such as ESMFold, a protein-structure model, to build containerized pipelines for bioinformatics and drug discovery. (aws.amazon.com) NVIDIA says Evo 2 NIM can run on cloud, data-center, or workstation infrastructure and is designed to simplify deployment for developers and system administrators. That packaging is part of a larger push by cloud vendors and chip companies to sell not just models, but managed ways to plug them into research and healthcare data systems. (docs.nvidia.com) There are limits alongside the launch. NVIDIA’s model card says Evo 2 was trained by Arc Institute rather than NVIDIA, and it places responsibility on users to comply with applicable laws when deploying the model. (docs.nvidia.com) So the announcement is a distribution move: a genome model released in February 2025 is now packaged for one-click use on Amazon’s machine-learning stack as NVIDIA tries to make biological artificial intelligence easier to run inside production pipelines. (aws.amazon.com)