Chip and circuit board design software — electronic design automation (EDA) — is set to run Nvidia built agents that re verify every design choice against Siemens' own tools and learn from past projects.
Siemens is wiring Nvidia-built AI agents into the chip and printed-circuit-board (PCB) design software engineers already use to lay out and verify semiconductors. Every design decision is now re-checked against Siemens' own verification tools, and the agents learn from past projects.
The product is called Fuse, and it is the new face of the "characterisation AI agent" Siemens unveiled at the Design Automation Conference (DAC) 2026. Fuse runs inside Siemens' Intelligence Center X and orchestrates the company's existing electronic design automation (EDA) tools. For a working chip or board, the agents drive the design loop: synthesise a candidate layout, hand it to Siemens' verification tools, then iterate against the results, with each run leaving context the next one can pick up.
The practical change is in how a designer's day goes. Instead of submitting a block of code to the synthesis tool, waiting on a verification run, and triaging failures by hand, the agent now handles that loop. Each candidate is fed through Siemens' timing, power, signoff, and test engines and re-tried until the result matches the constraints. "Self-verifying agents learn from every project to improve execution strategies and reuse accumulated context," Siemens said in its announcement.
The agents run on Nvidia's stack, which is what ties Siemens and Nvidia together. They are trained on Nvidia NeMo Gym, an open library for building agentic environments, and use Nvidia's Nemotron models and Switchyard to reason through engineering trade-offs. Nvidia OpenShell provides the secure runtime that lets a large enterprise audit and control the agents, and Nvidia's CUDA-X libraries (the company's accelerated-computing toolkit) are what Siemens says compresses signoff-quality timelines to "hours instead of day."
The new wiring is a deepening of the partnership Siemens and Nvidia first set out in March 2026, when Nvidia's Agent Toolkit, Nemotron models, and AI infrastructure were folded into Siemens' EDA automation workflows. The DAC announcement is the first time the collaboration has named concrete agent infrastructure rather than a general integration. Electronics Weekly reported that the joint claim is that the new EDA agents, built on NeMo Gym, improve "result quality, speed, and token efficiency."
Fuse's reach spans the Siemens EDA stack end to end. It is not a single tool but a layer that touches high-level synthesis (Catapult), verification (Questa One, Veloce), custom IC design and verification (Solido), physical implementation (Aprisa), signoff verification (Calibre), design-for-test (Tessent), 3D IC integration (Innovator3D IC), and PCB design (Xpedition). The "self-verifying" label is meaningful only because the substrate is in-house verification, not a generic large language model. The agent's job is to keep calling the right Siemens tool until the design passes.
The limits are real and worth naming. Every product and performance claim so far comes from Siemens and Nvidia. The "hours instead of day" signoff number is a vendor assertion, not an independent benchmark, and no customer, tape-out, or production chip has been named. The phrase "self-verifying" is also a marketing label: the underlying capability is the integration of Nvidia's agent infrastructure into an existing Siemens toolchain, not a new AI research result. The agents' accumulated project context is opaque until a customer references it.
Amit Gupta, senior vice president and chief AI strategy officer at Siemens EDA, has positioned the move as the next step in agentic EDA: tools that verify themselves rather than waiting on a human to run checks. The real test is whether the agent layer can keep calling the same Siemens verification tools a senior engineer would call, faster, and on a larger share of the design surface, without trading away the trust the tools were bought for. Siemens says a broader deployment of Fuse is in progress; the next signal to watch is whether any design house publishes an independent timing or signoff number on a production-class block.