Siemens and NVIDIA tout Agentic AI

Siemens has EDA tools and flows, NVIDIA has AI infrastructure, so what happens when these two companies collaborate? I’m at DAC 2026 and all the buzz this year is Agentic AI, so an came out on Sunday evening about how semiconductor and PCB design flows are being improved by this teamwork. Amit Gupta from Siemens and Timothy Costa from NVIDIA presented slides on Agentic AI and took some interactive questions last week during a video call.
Siemens has used AI technology within several EDA tools, like: Catapult for high-level synthesis, Calibre for physical verification, Solido for IC library characterization, Tessent for DFT, Questa One and Veloce for verification, Aprisa for physical implementation. In Siemens design and engineering tools, there is AI infrastructure for: Teamcenter, Simcenter, Designcenter, Digital Twin Composer. Industry trends are that AI accelerators are being embedded into chips, software-defined products drive hardware implementations, multi-die integration continues to grow, and power efficiency is a requirement. At Siemens they are bringing out AI-native tools and flows with faster engines, and smarter execution, all while providing deterministic results.
Last year Siemens launched Fuse EDA, an AI system for agentic semiconductor and PCB design that has been integrated into the new Intelligence Center X. This year in March at the NVIDIA GTC event we saw how the Fuse EDA AI Agent could plan, orchestrate and execute workflows across semiconductor and PBC Design by using the MCP architecture. At DAC the news is that there are now self-verifying and long-running agents that are using GPU-accelerated EDA tools, with human in the loop. NVIDIA NeMo Gym makes the agents run smarter, Nemotron models are used for token efficiency, all running under the OpenShell for secure operations.
Tim shared how the explosive transistor counts of systems reaching soon into the trillions require new methods like AI, as the old ways are just too slow. NVIDIA provides the Nemotron Open Models, data and software libraries to enable Agentic AI with Siemens tools. Over the years NVIDIA has provided CUDA-X libraries for a wide range of industries, like: Genomics, medical imaging, quantum computing, physics, deep learning, CAE, numerical computing, etc.
Siemens’ Intelligence Center X uses the CUDA-X libraries to support both agent creation and orchestration, extending this AI approach for companies that have design, manufacturing and supply chain operations. Using Digital Twins helps engineering teams make smarter execution and trusted outcomes.
During the Q&A time questions were answered about how agents work, trusting AI results, self-verification, EDA flows within NVIDIA, and where agents are used at Siemens.
Summary
DAC 2025 was all about using AI within EDA tools, and this year I see Agents controlling EDA tools, so the progress in just one year has been brisk. Siemens and NVIDIA have made agentic AI really work for semiconductor and PCB flows using deterministic, physics-based EDA engines. You can expect the domains that agentic AI address to build over time. Compressing design times from days to hours is happening with agentic AI like this.
Related Blogs
- GPU-native mask rule checking eliminates the curvilinear mask rule check bottleneck
- Library Characterization gets a Boost from AI
- Engineering the Next Era of Semiconductor Innovation
- DAC TechTalk – A Siemens and NVIDIA Perspective on Unlocking the Power of AI in EDA
Comments
There are no comments yet.
You must register or log in to view/post comments.

Facts Only

* Siemens uses AI in EDA tools including Catapult, Calibre, Solido, Tessent, Questa One, Veloce, and Aprisa.
* Siemens design and engineering tools include Teamcenter, Simcenter, Designcenter, and Digital Twin Composer.
* NVIDIA provides Nemotron Open Models, data, and software libraries to enable Agentic AI with Siemens tools.
* NVIDIA has provided CUDA-X libraries across various industries such as Genomics, medical imaging, quantum computing, physics, deep learning, CAE, and numerical computing.
* Siemens' Intelligence Center X uses CUDA-X libraries for agent creation and orchestration.
* Fuse EDA is an AI system integrated into the Intelligence Center X for agentic semiconductor and PCB design.
* The Fuse EDA AI Agent plans, orchestrates, and executes workflows using the MCP architecture.
* Agents are now self-verifying and long-running, utilizing GPU-accelerated EDA tools with human oversight.
* NVIDIA NeMo Gym and Nemotron models make agents run smarter, with Nemotron used for token efficiency under OpenShell.

Executive Summary

The collaboration between Siemens and NVIDIA focuses on integrating Agentic AI into semiconductor and PCB design flows. Siemens leverages AI within its existing EDA tools, such as Catapult, Calibre, Solido, Tessent, Questa One, Veloce, and Aprisa, alongside infrastructure like Teamcenter, Simcenter, Designcenter, and Digital Twin Composer. NVIDIA provides the necessary AI infrastructure through models like Nemotron and software libraries, including CUDA-X, to enable Agentic AI capabilities within these flows. This teamwork aims to address the challenges posed by rapidly increasing transistor counts and the need for faster design methods. A specific application demonstrated involves Fuse EDA, an AI system that plans, orchestrates, and executes workflows across semiconductor and PCB design using the MCP architecture at NVIDIA GTC events. The progression involves developing self-verifying, long-running agents utilizing GPU-accelerated EDA tools, enhanced by Nemotron models for token efficiency and secure operations under OpenShell.

Full Take

The dynamic described illustrates a critical shift from traditional, slow design methodologies to an agentic paradigm in hardware design, driven by the physical scaling of semiconductor systems. The core pattern involves coupling foundational physics-based execution (Siemens' EDA tools) with advanced intelligence layers (NVIDIA's AI infrastructure) to achieve workflow automation. This suggests that the bottleneck is no longer purely algorithmic but computational execution and orchestration across complex, multi-domain workflows. The integration point—using CUDA-X libraries within the Intelligence Center X framework—is significant; it establishes a shared substrate where physical verification and high-level planning can interact seamlessly under deterministic constraints. The implication is that future gains in design efficiency will depend not just on better EDA algorithms but on the ability to deploy intelligent, self-correcting agents capable of navigating multi-step engineering processes with guaranteed, physics-based fidelity. The pattern suggests a necessary synergy where specialized domain knowledge (Siemens' EDA) is generalized and accelerated by broad AI capabilities (NVIDIA's models), which challenges the assumption that pure simulation or pure deep learning must dominate this space. What is the cost structure for bringing these deterministic, agentic systems into widespread industry adoption? How will the concept of "human in the loop" scale when agents handle complex planning and execution across trillions of transistors?

Sentinel — Human

Confidence

The text reads like a summary or recap of a specific industry presentation, demonstrating deep subject knowledge and structure typical of human reporting, though it relies on synthesizing existing high-level concepts.

Signals Detected

Sentence length variance is present but generally flows logically; not overly uniform.

The text connects disparate concepts (EDA tools, AI infrastructure, specific company names) into a coherent narrative flow.

Uses standard reporting structures but the integration of specific product names suggests direct sourcing rather than pure aggregation.

Claims are heavily focused on connecting existing, real-world concepts (Siemens EDA, NVIDIA CUDA, Agentic AI) without introducing obvious factual errors or overly polished narrative that screams pure LLM generation.

Human Indicators

Inclusion of specific, current industry events (DAC 2026), named individuals (Amit Gupta, Timothy Costa), and proprietary tool names suggests direct engagement with specialized knowledge.

The tone is that of an industry commentator synthesizing technical presentations rather than generating theoretical concepts.