NVIDIA Factory Operations Blueprint, giving factories a new AI brain

NVIDIA has announced the NVIDIA Factory Operations Blueprint (FOX), a reference design for building autonomous factory management agents. FOX continuously monitors and reasons over real-time data, orchestrating specialized AI agents and devices responsible for quality control, material transport, and worker safety.

The blueprint is built on NVIDIA NemoClaw, AI-Q Blueprint, and NVIDIA Nemotron open models, and runs optimized on NVIDIA DGX Station. In particular, DGX Station equipped with the GB300 Grace Blackwell Ultra desktop superchip delivers the performance to run large-scale AI models with up to 1 trillion parameters locally.

Key features include the following:

• Connecting factory systems and agents: Integrates with industrial data sources, machines, and robot fleets, and connects with specialized agents through standard APIs.
• Automating AI model training: Uses NVIDIA TAO to automate the entire process from identifying accuracy gaps to data generation, model fine-tuning, and deployment.
• Running intelligent workflows: Enables visual inspection through NVIDIA Metropolis and building operational digital twins using NVIDIA Omniverse.

Taiwanese manufacturers including Advantech, Foxconn, Pegatron, and Wistron have been the first to adopt FOX. In particular, Foxconn, the world's largest electronics manufacturer, is building MoMClaw, a multi-agent system for manufacturing operations, using FOX and NemoClaw.

With MoMClaw featuring the security capabilities of NVIDIA OpenShell, Foxconn expects to cut root cause analysis time by 80%, improve labor productivity by 15%, and lower machine failure rates by 10%.