NVIDIA has expanded its NVIDIA Agent Toolkit by integrating NVIDIA PhysicsNeMo and CUDA-X libraries, marking a significant step toward autonomous AI-driven engineering. The update equips developers with specialized physics modeling, accelerated solvers, and quantum chemistry capabilities designed to streamline complex chip, system, and industrial design workflows. According to Timothy Costa, vice president and general manager of computational engineering at NVIDIA, the expansion enables agentic systems to reason using physics, execute high-fidelity simulations, and generate precise data, fundamentally shifting how engineers approach product development. Central to this update is NVIDIA Nemotron 3 Ultra, an open model that currently leads in agentic register-transfer level coding for chip design. The model supports post-training on proprietary datasets and can be deployed locally or on-premises, providing enterprises with enhanced data privacy and customization. NVIDIA has aligned these tools with major industry players to accelerate autonomous design pipelines. Cadence has integrated Nemotron and CUDA-X into its AuraStack AI Super Agent platform, achieving up to a twentyfold improvement in multiphysics performance for advanced packaging and PCB design. Synopsys is leveraging the toolkit alongside Ansys Icepak to autonomously handle GPU cooling optimization, while also optimizing its VCS verification platform for NVIDIA Vera CPUs. Siemens utilizes NeMo Gym and CUDA-X with its Fuse EDA AI Agent to orchestrate multi-tool semiconductor workflows, delivering over a tenfold acceleration in library characterization. Additional enterprise adopters include Samsung, which applies cuLitho and PhysicsNeMo for computational lithography and chip-scale thermal-stress analysis, yielding up to twentyfold performance gains. Silvaco reports completing a 3.2-billion-mesh-node photonic simulation in under four hours using thirty-two NVIDIA GPUs, while Keysight harnesses cuDSS for electromagnetic simulations that run up to ten times faster. Samsung, Synopsys, and TSMC are integrating cuEST to achieve up to a fiftyfold speedup in quantum-chemistry workloads. ChipAgents is fine-tuning Nemotron models for end-to-end semiconductor verification. The expanded toolkit and partner implementations will be highlighted at the upcoming Design Automation Conference, underscoring NVIDIA's strategy to embed AI-driven automation directly into foundational engineering infrastructure.