NVIDIA Announces Blueprint for Open Physical AI Data Factory

NVIDIA has introduced the Physical AI Data Factory Blueprint, an open reference architecture designed to standardize and automate how training data is generated and evaluated for physical AI systems. The blueprint targets robotics, vision-based AI agents, and autonomous vehicle development, with an emphasis on scaling data pipelines that combine real-world data, simulation, and synthetic data generation. This approach aligns with recent ARC Advisory Group research on physical intelligence, which highlights data availability and quality as one of the largest limiting factors in advancing autonomous and agentic systems in industrial domains.

The blueprint integrates large-scale data curation, synthetic data generation, reinforcement learning workflows, and automated evaluation into a unified pipeline. By using simulation and world foundation models to expand limited datasets, the architecture addresses the challenge of capturing rare and long-tail scenarios that are difficult or costly to collect in physical environments with real data.

NVIDIA is working with cloud providers, including Microsoft Azure and Nebius, to make the blueprint deployable as a cloud-based data production environment. These integrations are intended to convert large-scale compute resources into repeatable, agent-driven data pipelines that support training and validation workflows.

The architecture also incorporates agent-driven orchestration through NVIDIA’s open-source OSMO framework, which coordinates data generation, augmentation, and evaluation across heterogeneous compute environments. This focus on orchestration mirrors ARC’s view that future physical AI platforms will differentiate less on individual models and more on their ability to manage complex, distributed pipelines spanning simulation, data management, and deployment. These orchestration layers are increasingly central to scaling autonomous systems beyond pilots into production environments.

The Physical AI Data Factory Blueprint is expected to be released in April.

For more coverage of Physical Intelligence, please visit our website.