NVIDIA Expands Agent Toolkit
NVIDIA has announced a significant expansion of its NVIDIA Agent Toolkit, introducing new agent-ready capabilities through NVIDIA PhysicsNeMo and enhanced CUDA-X libraries.
The update is designed to help developers build AI agents that support complex engineering tasks, including semiconductor design, simulation and verification.
The announcement reflects NVIDIA’s continued investment in agentic AI for engineering. According to the company, developers can use the toolkit to create specialised AI agents that interact with engineering software, automate technical workflows and accelerate scientific computing.
New Capabilities for Engineering AI
The expanded NVIDIA Agent Toolkit now includes access to PhysicsNeMo, NVIDIA’s framework for physics-informed AI, alongside an expanded collection of CUDA-X libraries for accelerated engineering workloads.
Together, these additions allow AI agents to move beyond natural language tasks. They can incorporate engineering knowledge, physics-based reasoning and high-performance numerical computing into technical workflows.
The toolkit is intended for applications across semiconductor engineering, manufacturing, robotics, materials science and other computationally intensive industries.
Support for Semiconductor Design and Verification
One of the key announcements for the semiconductor industry is NVIDIA’s focus on AI-assisted RTL development and verification.
The company said its Nemotron 3 Ultra model, combined with the ACE-RTL agent developed by NVIDIA Research, delivers leading benchmark performance for agentic register-transfer level (RTL) coding among open models.
According to NVIDIA, enterprises can use these technologies to build custom AI agents for chip design and verification while maintaining control over proprietary engineering data.
Potential applications include RTL development, verification workflows, design analysis and engineering automation throughout the semiconductor development lifecycle.
PhysicsNeMo Brings Physics-Based AI to Engineering Workflows
AAs part of the update, NVIDIA has restructured PhysicsNeMo into modular, agent-ready components that integrate directly into AI engineering workflows.
PhysicsNeMo combines machine learning with physical modelling, enabling AI agents to assist with simulation-driven engineering tasks. These workloads have traditionally required significant computational resources.
NVIDIA says the framework can accelerate modelling across multiple engineering disciplines while maintaining physics-informed accuracy.
The company has also expanded access to CUDA-X libraries. This gives AI agents access to accelerated numerical solvers, optimisation libraries and scientific computing tools for engineering workloads.
Looking Ahead
Alongside the Agent Toolkit expansion, NVIDIA highlighted growing adoption across the electronic design automation (EDA) industry.
Companies including Cadence, Synopsys, Siemens and ChipAgents are already integrating NVIDIA’s AI models, accelerated computing technologies and agent frameworks into semiconductor design, verification and simulation workflows.
These collaborations span applications including AI-assisted PCB and advanced packaging design, formal verification, RTL development, thermal analysis, computational lithography and electromagnetic simulation.
Several companies also reported notable performance gains:
- Up to 20x faster multiphysics performance
- 10x faster library characterization
- 10x reduction in token costs
- 20x greater performance for computational lithography
- 10x accelerattion of electromagnetic simulations
As adoption grows, AI agents are expected to play an increasingly important role across the semiconductor design and verification lifecycle. Their role will be to complement engineers by improving automation, accelerating simulation and streamlining complex engineering workflows.