Siemens possesses a complete set of EDA tools and design processes, while NVIDIA provides a full suite of AI 

computing infrastructure. What kind of chemical reaction will occur when these two collaborate? At this year's 

Design Automation Conference (DAC 2026), the topic of proxy AI was widely discussed. On Sunday evening, both 

parties released related special content, elaborating on how this collaboration optimizes the entire design 

process of semiconductors and printed circuit boards (PCBs). Last week, Siemens' Amit Gupta and NVIDIA's 

Timothy Costa conducted an online presentation, sharing slides on proxy AI-related technologies and answering 

audience interaction questions online.

Siemens has already implemented AI technology in several mainstream EDA tools, including: Catapult for high-

level synthesis, Cal for physical verification, Solido for IC library characterization, Tessent for testability design 

(DFT), Questa One for simulation and verification, Veloce for hardware simulators, and Aprisa for layout and 

routing. At the same time, all of Siemens' complete design engineering software is equipped with an AI 

foundation, covering the product lifecycle management platform Teamcenter, the multi-physics simulation suite 

Simcenter, the design center Designcenter, and the digital twin construction tool Digital Twin Composer.

The current mainstream industry trend: AI accelerators are gradually embedded inside chips, software-defined 

hardware products are increasingly popular, multi-chip integration scales are continuously expanding, and energy 

efficiency has become a hard design metric. Siemens is launching native AI tools and processes, with faster 

computing engines and stronger intelligent scheduling capabilities, while ensuring that simulation and design 

results are deterministic.

Siemens launched the Fuse EDA AI system last year. This platform has been integrated into the new intelligent 

center X (Intelligence Center X), specifically designed for the semiconductor and PCB fields, providing a proxy AI 

full-chain solution. In March at the NVIDIA GTC Global Technology Conference, the industry witnessed the Fuse 

EDA intelligent body autonomously planning, scheduling, and executing the entire design workflow for 

semiconductors and PCBs based on the MCP architecture. At this year's DAC conference, the latest progress was 

disclosed: an intelligent body now supports GPU-accelerated EDA tools, can operate stably for a long time, and 

has self-checking capabilities. The entire process adopts a human-machine collaborative mode. NVIDIA's NeMo 

intelligent training environment can enhance the running logic of the intelligent body, Nemotron large model 

optimizes token encoding efficiency, and the entire system executes various tasks under the OpenShell secure 

running framework.

Tim shared his views: The transistor scale of future chip systems will approach the trillion-level, and traditional 

design methods are severely inefficient. We must rely on new technical paradigms like AI to break through. 

NVIDIA opens the Nemotron open-source large model, supporting datasets and software libraries, to support 

the implementation of proxy AI. Over the years, NVIDIA has launched CUDA-X acceleration function libraries for 

various industries such as genomics, medical imaging, quantum computing, physical simulation, deep learning, 

computer-aided engineering (CA), and numerical computing. 

The Siemens intelligent center X (Intelligence Center X) calls on the CUDA-X acceleration library to achieve two 

core capabilities of intelligent body creation and process coordination, and this AI solution can cover the entire 

industrial chain enterprises with design, manufacturing, and supply chain businesses. Digital twin technology 

assists the engineering team in formulating more reasonable implementation plans and outputs reliable design 

results.

The Q&A session focused on topics such as the operation principle of intelligent agents, the credibility of AI 

results, the autonomous verification mechanism, Nvidia's internal EDA process, and the implementation scenarios 

of intelligent agents in each business line of Siemens. Summary

The 2025 DAC conference focuses on "Embedded AI Assistance Functions in EDA Tools". This year, the industry 

has entered a new stage of autonomous operation of EDA processes by intelligent agents. The technological 

iteration speed in just one year has been extremely rapid. Siemens and NVIDIA have successfully implemented 

an agent-based AI solution that adapts to the entire semiconductor and PCB production chain based on the EDA 

underlying engine with deterministic physics. The coverage of design scenarios for agent-based AI will continue 

to expand. With this technology, design tasks that originally took several days can be completed in just a few 

hours.