Nvidia launches USD BRep geometry kernel for solid modelling

AI-generated summary

Nvidia has launched USD BRep, a standalone C++ and Python geometry library for NURBS-based boundary representation solid modelling, with native import and export of BRep data to and from OpenUSD. The article reports that the release is prompting developers to explore new open-source CAD software designed around AI tools and workflows. Its broader significance, however, may lie in connecting precise engineering geometry with visualisation, industrial digital twins and downstream applications rather than replacing established CAD systems. The AI-to-CAD problem addressed by USD BRep is the difficulty of training machine-learning algorithms on traditional CAD data, including associated metadata and multistep feature trees. The kernel exposes NURBS and BRep geometry directly through Python and C++ structures, giving developers access to exact geometric boundaries for neural-network training and generative-model development. The article presents this as a way to support large language models and diffusion networks in creating mathematically sound, manufacturable 3D models, although it does not provide examples or performance results demonstrating those outcomes. BRep represents exact three-dimensional solids through their vertices, edges and faces, while NURBS—Non-Uniform Rational B-Splines—describe smooth mathematical curves and surfaces, including cylinders and organic shapes. Unlike polygon meshes, these representations preserve the precise geometry required for engineering work. USD BRep allows that engineering data to reside natively within USD layers, simplifying OpenUSD-based visualisation and VR/XR workflows while creating opportunities for new modelling applications. Combined with Nvidia’s Omniverse platform and its framework for industrial digital twins, this integration holds the promise of enabling engineers to manipulate large industrial assets in real time without sacrificing dimensional accuracy. The article also highlights the potential to combine precise CAD geometry with the photorealistic visualisation associated with OpenUSD digital twins. These are presented as opportunities enabled by the architecture, rather than established results from a documented industrial implementation. The article nevertheless argues that established CAD kernels are unlikely to be displaced quickly. Siemens’ Parasolid has more than 30 years of development behind it, including troubleshooting and resolution of mathematical edge cases. Its use across Siemens software, Solidworks, Onshape and other products gives it an entrenched industry position that the author expects to remain important for years. Siemens Digital Industries CEO Tony Hemmelgram offers a contrasting view of AI-driven CAD: the main value is not another modeller or free kernel, but tighter connections between capable existing CAD tools and language models. He points to the APIs in NX and Designcenter and cites a project linking NX with Claude as an example of an LLM driving an established engineering application. USD BRep also builds on an existing geometry technology. The open-source project includes SMLib, the Solid Modeling Library originally developed by Solid Modeling Solutions, which Nvidia acquired in May 2022. SMLib currently powers geometry processing and solid modelling in AWV Informatik’s ClassCAD. AWV Informatik CEO Weigel Rainer, whose company has worked with the SMLib team for more than 20 years, welcomed the kernel becoming open source and expressed interest in what the wider CAD community would build with it. The author concludes that Nvidia appears more focused on improving engineering-data pipelines, translation and downstream infrastructure than on replacing industry-standard CAD tools. The immediate opportunity is to move complex BRep geometry into applications such as Omniverse without reducing it to polygon meshes and losing its precise mathematical representation. As a related development, the article notes that Unity has released a USD Toolkit in partnership with Nvidia, bringing OpenUSD support to Unity and its visual digital twins.