NVIDIA pushes an ‘always-on’ factory-agent blueprint across Europe’s manufacturing ecosystem

NVIDIA says Europe’s industrial leaders are using its AI and simulation software stack to redesign engineering and factory operations, with an emphasis on moving from one-off analytics toward coordinated, “always-on” systems that can monitor production in real time and take structured next steps. The company highlighted a Factory Operations Blueprint intended to give developers a path to build agents that can coordinate across machines and specialist tasks across the manufacturing lifecycle.

At the core of the effort is NVIDIA’s Factory Operations Blueprint, or FOX, which NVIDIA says is available in early access on GitHub. NVIDIA describes FOX as a framework for developers to build and deploy factory-manager agents that coordinate work across machines, along with specialist agents for specific workflows such as visual inspection, root cause analysis, report generation, and fleet management.

NVIDIA’s blueprint is designed to connect industrial agents to a range of models and engineering data. The company says the agents can connect to “frontier models,” including NVIDIA Cosmos and Nemotron, as well as foundational knowledge such as computer-aided design, product lifecycle management, simulation software, and IT and operational technology systems. NVIDIA also says its approach relies on Cosmos in-context learning, meaning the agents can adapt using examples and documentation without requiring fine-tuning.

NVIDIA’s blog also points to integration work with Siemens, describing how Siemens is incorporating FOX-built agents into its Intelligence Center X, an agentic enterprise system aimed at coordinating enterprise workflows. NVIDIA says the combined approach is estimated to deliver 20% cost savings, tied in part to reducing engineering and production friction through faster diagnosis and coordinated actions on the factory floor.

In Siemens’ “Process Preparation X” workflow, NVIDIA says the use of NVIDIA Cosmos Data Factory tools and skills is aimed at generating defect images for circuit-board inspection. NVIDIA adds that at Siemens’ Erlangen electronics factory, the company used the approach to cut calibration time for inspection processes by 40%. While NVIDIA does not provide additional methodological detail in the release, it frames the results as evidence that better preparation and agent-driven coordination can reduce time spent tuning industrial inspection systems.

The release ties FOX and industrial agents to NVIDIA’s broader simulation push. NVIDIA says its CUDA-X libraries accelerate simulation so engineers can get faster feedback while exploring design alternatives before building physical prototypes. It also highlights a GTC Berlin release of cuDSS for general availability, describing cuDSS as software intended to accelerate demanding calculations used in structural stress, fluid flow, electromagnetic fields, and semiconductor device modeling.

NVIDIA’s examples extend across automotive, semiconductors, aerospace, and industrial robotics. Mercedes-Benz is cited as using Siemens’ Simcenter STAR-CCM+ running on NVIDIA GPUs to compare vehicle designs through faster aerodynamic simulations, and NVIDIA says Mercedes uses Cosmos 3 and a Metropolis VSS blueprint for vehicle inspection and windshield assembly verification. In semiconductors, NVIDIA points to imec and Infineon using Synopsys software alongside NVIDIA-accelerated computing, including benchmarks for S-Litho at up to 40x speedup and a 12x speedup for 3D breakdown simulation using Infineon’s Sentaurus Device with NVIDIA CUDA libraries, according to the release.

Other named efforts include Leonardo working with Neural Concept on AI-driven aircraft design using NVIDIA Omniverse libraries to connect aircraft geometry and aerodynamic simulation results; Dassault Systèmes leveraging Nemotron-powered “LEO & AURA Virtual Companions” and orchestrating CATIA AI Agents and SIMULIA AI Agents with SIMULIA solvers accelerated by CUDA-X libraries; and UniversalAGI’s work with Saab that uses physics foundation models to evaluate thousands of wave conditions and headings in real time to help anticipate dangerous wave motion at sea.

The company also positions digital twins and robotics simulation as a parallel pathway for industrial change. NVIDIA cites use cases such as factory concept validation with Omniverse (including an Audi battery-production digital twin), humanoid actuator digital twins using NVIDIA Isaac simulation frameworks (at Schaeffler), and assembly workflow validation using Isaac Sim and Omniverse/OpenUSD (via Wandelbots). In each example, the common theme is shifting validation and iteration earlier in the process by running realistic simulations and agent-driven “what-if” scenarios before making physical changes.

NVIDIA’s announcement does not provide broader performance benchmarks for FOX itself, nor does it spell out deployment timelines, data security standards, or regulatory considerations for production environments. It also does not clarify how different vendors’ machine-control systems and software stacks would be standardized for agent coordination beyond the stated ability to connect to IT and operational technology systems. For developers and industrial buyers, those gaps may be the key next questions to resolve as early access expands beyond initial partners.

For what to watch next, NVIDIA’s FOX early-access availability on GitHub suggests the company wants to draw in more industrial developers and systems integrators, not just enterprise customers. In parallel, the release points to a wider ecosystem of simulation libraries and agent platforms, meaning competition and consolidation may shift toward who can operationalize simulation-informed AI inside factories, inspection pipelines, and enterprise coordination layers. The next milestones likely include public documentation updates for FOX, more partner case studies with verified outcomes, and additional interoperability details connecting agents to specific industrial toolchains.