NVIDIA Launches Ising: First Open AI Models for Quantum Computing
Open reference architecture uses Cosmos models and AI agents to generate massive, rare-scenario datasets.
NVIDIA has unveiled the Physical AI Data Factory Blueprint, an open reference architecture designed to unify and automate the creation of training data for physical AI systems like robotics, vision AI agents, and autonomous vehicles. The core challenge in this domain is generating the massive, diverse datasets—including expensive-to-capture rare edge cases—required for AI to operate reliably in the real world. NVIDIA's solution leverages its Cosmos suite of open world foundation models (Curator, Transfer, Evaluator) within a modular workflow to curate, exponentially augment, and validate data. Crucially, the blueprint integrates with leading coding agents like Claude Code and OpenAI Codex via the NVIDIA OSMO orchestration framework, enabling AI-native, agent-driven management of the entire data generation pipeline at scale.
NVIDIA is collaborating with cloud giants Microsoft Azure and Nebius to integrate this blueprint directly into their infrastructure, allowing developers to transform raw compute power into high-volume, high-quality training data. Early adopters include major players like Skild AI, which is using it to advance general-purpose robot foundation models, and Uber, for accelerating autonomous vehicle development. NVIDIA itself is using the blueprint to train its Alpamayo model, touted as the world's first open reasoning-based vision language action model for autonomous driving. By providing this standardized, agent-orchestrated 'data factory,' NVIDIA aims to drastically reduce the cost, time, and complexity of developing physical AI, essentially arguing that in this new era, 'compute is data.'
- Automates data pipeline: The blueprint provides a unified architecture using NVIDIA Cosmos models (Curator, Transfer, Evaluator) to automate curation, augmentation, and validation of training data for physical AI.
- Agent-driven orchestration: Integrates with coding agents (Claude Code, OpenAI Codex) via the NVIDIA OSMO framework to manage complex AI infrastructure and workflows at scale, reducing manual tasks.
- Cloud & industry adoption: Partnered with Microsoft Azure and Nebius for cloud integration; already used by Skild AI for robots, Uber for AVs, and NVIDIA for its Alpamayo driving model.
Why It Matters
It solves the critical data bottleneck for real-world AI, letting teams generate rare-scenario training data at scale to build safer robots and autonomous vehicles faster.