AU 2025: Autodesk announces new 3D generative AI foundation models for Fusion and Forma with neural CAD and AutoConstrain features

At Autodesk University 2025 (AU 2025), Autodesk unveiled its upcoming 3D generative AI foundation models for Fusion and Forma, marking a significant step forward in design and make technology. These models, described as neural CAD, are designed to assist designers and architects in seamlessly transitioning from early-stage concepts to detailed designs. The announcement positions Autodesk at the forefront of integrating artificial intelligence directly into CAD workflows, transforming traditional approaches that have remained largely unchanged for decades.

Neural CAD for generative design
The core feature of these foundation models is neural CAD, a generative AI model capable of reasoning directly about CAD objects as well as industrial and architectural systems. Unlike conventional parametric CAD engines, which have dominated design for over 40 years, neural CAD can generate designs spontaneously based on text prompts. For AEC applications, this allows users to quickly move from conceptual sketches or rough layouts to detailed building plans and system designs. By leveraging decades of project knowledge, neural CAD supports iterative exploration, helping teams optimize cost, sustainability, and project suitability early in the design process.

AutoConstrain for workflow efficiency
Another feature highlighted at AU 2025 is the enhanced AutoConstrain tool in Fusion. First launched seven months ago, AutoConstrain assists designers by automatically applying constraints to their drawings, significantly reducing manual effort and streamlining workflows. The integration of AutoConstrain with neural CAD foundation models enhances the efficiency of both early-stage concept generation and detailed design refinement. Together, these features exemplify Autodesk’s approach of blending machine learning with CAD to create more intuitive, collaborative, and productive design experiences.

From research to application
Autodesk’s AI Lab, established in 2018, has been instrumental in developing these neural foundation models. Researchers focused on creating AI capable of understanding geometry, systems, and real-world constraints. The models are trained not only to handle detailed CAD geometry but also to reason about industrial processes, such as efficient machine tool paths and standard building layouts. Future iterations of these models will allow customization based on an organization’s proprietary data and workflows, further expanding their applicability across different industries.