NVIDIA has significantly expanded its Agent Toolkit for engineering workflows, integrating new NVIDIA PhysicsNeMo and CUDA-X libraries to enable the development of autonomous AI engineers. This crucial update, announced on 27 July 2026, aims to redefine engineering, design, and construction processes by allowing AI to reason with physics, run complex simulations, and generate high-fidelity data.
The move builds on the toolkit’s initial launch on 16 March 2026, at GTC 2026, and a subsequent expansion on 20 July 2026, which incorporated NVIDIA Omniverse libraries. This progression underlines AI’s growing role as a transformative force in product design and development.
Empowering engineering AI assistants with advanced tools
NVIDIA’s latest additions allow developers to construct sophisticated AI assistants that can interact directly with domain-specific tools, models, and data. This deeper integration of AI into critical engineering stages promises to tackle challenges previously requiring extensive human intervention.
Timothy Costa, vice president and general manager of computational engineering at NVIDIA, noted that engineering has reached an inflection point. He stated that AI can now work with tools of physics, simulation, and design, enabling developers to build agentic engineers that reason using physics, run complex simulations, and generate high-fidelity data.
PhysicsNeMo and CUDA-X libraries enhance capabilities
The re-architected PhysicsNeMo libraries now provide advanced AI physics skills, allowing agents to train and deploy customizable AI physics models for complex design and simulation tasks. These libraries effectively translate intricate model architectures into callable tools, making physics-based reasoning readily accessible within engineering workflows.
Alongside PhysicsNeMo, the updated CUDA-X libraries infuse agentic engineering with accelerated solvers and quantum chemistry functionalities. This includes the new NVIDIA cuISS (CUDA Iterative Sparse Solvers) library, which accelerates large sparse linear systems used in physics-based and engineering simulations. Its composable solvers and preconditioners help developers build robust GPU-based simulation engines.
Furthermore, NVIDIA cuDSS (CUDA Direct Sparse Solvers) accelerates large sparse linear systems in electronic design automation and scientific simulation. It supports workloads like device, circuit, and system simulations, offering scalability to multi-GPU and multi-node deployments. For advanced material science, NVIDIA cuEST (CUDA Electronic Structure Theory) facilitates quantum chemistry simulations, integrating density functional theory and post-DFT methods into GPU-accelerated production workflows.
Streamlining chip design with agentic AI
Chip design relies heavily on register-transfer level (RTL) coding, demanding precision and extensive domain knowledge. This expansion directly addresses these challenges, pushing the boundaries of what AI can achieve in hardware development.
NVIDIA Nemotron 3 Ultra, powered by ACE-RTL from NVIDIA Research, ranks among leading open models for agentic RTL coding on the Verilog design problems benchmark. This represents a significant step forward in automating what has traditionally been a highly manual process.
Nemotron 3 Ultra for RTL coding efficiency and privacy
The ability to post-train Nemotron 3 Ultra on proprietary data, coupled with flexible deployment options—either locally or on-premises—is a critical advantage. This ensures enterprises maintain precise control over customisation, data privacy, and security when developing AI agents for their specific chip design needs.
Major industry players such as Cadence, Synopsys, and Siemens are already leveraging Nemotron 3 Ultra. They’re integrating it into their design verification and analog/mixed-signal workflows. Siemens’ Questa One smart verification agentic toolkit also benefits directly from this integration, showing broad industry adoption. Siemens and Nvidia are building next-gen industrial AI operating systems together, illustrating their collaborative efforts.
Overcoming engineering challenges through accelerated computing
The increasing complexity of modern engineering problems has pushed traditional methods to their limits. Engineers constantly grapple with balancing performance, cost, and time-to-market. The NVIDIA Agent Toolkit directly confronts these issues by providing scalable, intelligent automation that redefines conventional workflows.
This capability effectively turns AI into a force multiplier for human ingenuity. It allows engineers to explore a far wider design space and identify optimal solutions much faster. This will free up human engineers for more strategic and creative tasks.
Quantifiable performance gains and cost reduction
The reported performance gains from these toolkit expansions are substantial, directly impacting engineering productivity and project timelines. Multiphysics performance, crucial for simulating complex physical phenomena, sees an acceleration of up to 20 times. Library characterisation, essential for chip design, is now 10 times faster.
Engineers will find token costs reduced by a factor of 10, making extensive AI model use more economically viable. Computational lithography processes now demonstrate 20 times greater performance, while electromagnetic simulations are 10 times faster. Crucially, RTL verification can be compressed from weeks into mere hours.
The innovative AI-Q Blueprint architecture further reduces query costs by over 50%. Nemotron 3 Ultra also delivers 5x faster inference and up to 30% lower cost for complex agentic tasks. These efficiencies translate into tangible savings in both development time and financial outlay.
Enhanced security and accessibility for developers
Deployment challenges for AI agents often revolve around cost, security, and achieving true autonomy. NVIDIA addresses the security aspect with NVIDIA OpenShell, an open-source runtime that creates a secure, isolated sandbox for each agent.
This ensures agents cannot access sensitive files beyond their defined policy, providing essential peace of mind for enterprises handling critical intellectual property. It’s a vital component for fostering trust in AI-driven engineering. Industrial connectivity can also benefit from robust security measures.
Availability and industry adoption
The uptake of these technologies by major industrial software providers underscores their growing significance. Companies like Cadence, Siemens, Synopsys, SideFX, and PTC are already integrating NVIDIA’s accelerated computing and agentic AI into their offerings.
Neil Barua, President and CEO of PTC, highlighted the growing demand for connected design, collaboration, and simulation throughout the development process. He noted that NVIDIA Omniverse libraries facilitate simulation-ready workflows, bringing validation and testing closer to the product design phase. This collaboration ensures AI agents operate seamlessly across diverse software environments.
NVIDIA is also making these powerful tools widely accessible. The Nemotron 3 Ultra model is set for broader availability around June 4 (2026 implied) through platforms like Hugging Face, ModelScope, and NVIDIA NIM microservices. This widespread deployment will democratise access to advanced AI for chip design.
New RTX Spark Systems will launch in Fall (2026 implied) from ASUS, Dell Technologies, HP, Lenovo, Microsoft Surface, and MSI, with Acer and GIGABYTE models to follow. DGX Station Systems are also available to order from ASUS, Dell, GIGABYTE, HP, MSI, Supermicro, and Exxact. Production stability in energy, for instance, can be greatly enhanced through these advanced engineering tools, showcasing their broad applicability.
