NVIDIA (NVDA) Expands Toolkit With PhysicsNeMo and CUDA-X
NVIDIA (NASDAQ: NVDA) has announced a major update to its NVIDIA Agent Toolkit, integrating NVIDIA PhysicsNeMo and CUDA-X libraries to enable the development of autonomous AI engineers. These enhancements are designed to transform complex engineering workflows, particularly in chip design and system-level simulations.
PhysicsNeMo, an open-source framework for AI-driven physics simulations, allows for the training and deployment of customized models across applications like computational fluid dynamics and large-scale digital twins. CUDA-X, NVIDIA’s GPU-accelerated library stack, brings advanced solvers, quantum chemistry capabilities, and scalable multi-GPU performance to the table. Together, these tools aim to streamline engineering workflows by combining AI physics models with accelerated computing power.
"Engineering has reached an inflection point," said Timothy Costa, vice president of computational engineering at NVIDIA. "With NVIDIA Agent Toolkit, developers can build AI engineers capable of running simulations, reasoning with physics, and generating high-fidelity data to drive innovation in chip and system design."
Key additions include:
- PhysicsNeMo Libraries: AI physics skills for training and deploying models in engineering workflows.
- CUDA-X Updates: Enhanced libraries like cuISS and cuDSS for solving large-scale, sparse systems central to chip and system simulations.
- Quantum Chemistry: NVIDIA cuEST enables scalable quantum-chemistry simulations for semiconductor development.
Major industry players, including Cadence, Synopsys, and Siemens, are already leveraging these tools. Cadence has reported up to 20x faster multiphysics performance for packaging and PCB design workflows, while Siemens has achieved 10x faster library characterization using agentic AI workflows. Additionally, Samsung is utilizing PhysicsNeMo and CUDA-X for chip-scale thermal-stress analysis and computational lithography, delivering up to 50x speed improvements in certain tasks.
The announcement follows a series of strategic moves by NVIDIA to solidify its dominance in AI and accelerated computing. On June 8, 2026, NVIDIA partnered with SK hynix to integrate CUDA-X and PhysicsNeMo into semiconductor simulations, further underscoring the growing adoption of its tools across the tech industry. As of July 24, 2026, NVIDIA's market cap stands at $5.05 trillion, with shares trading at $206.84, reflecting its position as a leader in AI-driven engineering solutions.
For developers and enterprises looking to adopt these tools, NVIDIA offers access to Nemotron 3 Ultra, an open model tailored for agentic register-transfer level (RTL) coding. It supports proprietary data training and local deployment, ensuring customization and data privacy. Companies like Synopsys have integrated this model into their AI workflows for chip design and verification, optimizing performance and cutting simulation times significantly.
By expanding its toolkit, NVIDIA is positioning itself at the forefront of a growing trend: the use of AI to automate and accelerate complex engineering tasks. With PhysicsNeMo and CUDA-X, NVIDIA is not only enabling faster, more efficient design cycles but also redefining what’s possible in chip design, industrial simulation, and beyond.
To explore these advancements, NVIDIA will showcase its expanded toolkit at the Design Automation Conference (DAC) later this week, a key event for those in semiconductor design and AI engineering.