NVIDIA’s Open Data Initiative

NVIDIA is committed to building open data for AI, recognizing that high-quality datasets are essential for training reliable models. The company has released over 2 petabytes of AI-ready training data across 180 datasets and 650+ open models.

AI-Data Bottlenecks

Building high-quality datasets remains a significant challenge in AI development. NVIDIA aims to reduce this friction by publishing permissively licensed datasets on HuggingFace, along with training recipes and evaluation frameworks on GitHub.

For example, the Physical AI Collection includes 500K+ robotics trajectories, 57M grasps, and 15TB of multimodal data. This collection has been downloaded over 10 million times and has supported the development of models such as the NVIDIA GR00T reasoning vision-language-action model.

Real-World Open Datasets

NVIDIA’s open data releases span multiple domains, including robotics, autonomous systems, sovereign AI, biology, and evaluation benchmarks. The Nemotron Personas Collection, for instance, provides fully synthetic persona datasets grounded in real-world demographic distributions, supporting Sovereign AI development.

Additionally, the La Proteina dataset is a fully synthetic, atomistic protein dataset designed for biological modeling and drug discovery workflows. The SPEED-Bench is a standardized benchmark for evaluating speculative decoding performance, featuring two splits: a Qualitative Split and a Throughput Split.

Benefits of Open Data

Open data access gives developers a faster and more cost-effective path to building high-quality models, while making evaluation and improvement easier across the ecosystem. NVIDIA’s open data initiative has already supported the development of state-of-the-art models and has been adopted by companies such as CrowdStrike and NTT Data.