- Siemens is pitching open EDA as its edge in agentic chip design, letting engineers mix its tools with rival software from Synopsys and Cadence.
- Behind the pitch, adoption is early: 20-plus customers have installed its Fuse system, but only a handful have moved it into production.
Siemens is betting that open EDA, the freedom to mix and match chip-design tools across competing vendors, is the wedge that sets it apart in a market where every major player is now selling the same agentic AI promise. It is a strategy born as much from necessity as conviction, and one of the company’s most senior chip-software executives is candid about why.
Speaking on the sidelines of Realize LIVE Asia-Pacific in Bengaluru, Ankur Gupta, executive vice president of EDA IC software at Siemens Digital Industries Software, was direct about the problem facing every electronic design automation (EDA) vendor. “They all address a common problem,” he said. “Engineers want to get agentic AI help to explore the design space. Because now that these capabilities are there, everybody wants to use them.”
At NVIDIA’s GTC in March, Siemens launched its Fuse EDA AI Agent; within weeks Synopsys and Cadence had rolled out their own autonomous design agents, and by the Design Automation Conference in July all three were pitching long-running autonomy. On press releases, as Gupta acknowledges, the announcements read almost interchangeably. The harder question, the one readers actually want answered, is what separates them once the marketing is stripped away.
Why open EDA is a strategy, not a slogan
Gupta’s answer is structural, and it starts with an admission most vendors would avoid. Unlike Synopsys and Cadence, Siemens does not have a dominant, fully utilised platform that customers run end to end. That gap, he argues, is precisely why openness works in its favour.
“The biggest difference is Fuse is open,” he said. “What we mean in EDA by open is that you can mix and match tools. This is in the DNA of Siemens, especially Siemens EDA. We are well known for a few products, and we are building up a full platform. But unlike Cadence and Synopsys, we don’t have a highly utilised full platform. So because of that, we have to be open, because we have to be in a mix-and-match flow.”
In practice, that means a customer can pair a timing solution from one vendor, a place-and-route tool from another, and Siemens’ own physical verification or test software, then let Fuse orchestrate across all of it, provided each tool exposes well-defined APIs or Model Context Protocol (MCP) servers for the agents to call. Siemens’ own materials describe Fuse as a “true multi-vendor tool solution” running on an open framework for third-party integration.
The competitive contrast is real. Synopsys has built its agentic push, AgentEngineer, on the multi-physics stack it gained through its Ansys acquisition, positioning a tool-native approach as the more controllable one. Cadence has assembled its own agent portfolio across its integrated flow. Both are betting that autonomy is safest when the agent lives inside a single vendor’s tightly held stack. Siemens is betting the opposite, that customers will not, and increasingly cannot, commit their entire flow to one supplier.
Gupta claims the door only swings one way. “If you were in one of those systems, our understanding is it’s not as easy to do the reverse,” he said. Fuse, he argues, is designed from the outset to reach across vendor lines rather than fence customers in.
The adoption reality behind the open EDA pitch
Where Gupta’s account earns credibility is that he does not oversell the traction. Asked how far Fuse has actually travelled, he laid out an adoption ladder rather than a headline.
“The installations that we are doing with customers, that’s 20-plus,” he said. Getting there is not instant: setting Fuse up to work with a customer’s on-premise compute, its GPUs, CPUs and scheduling, “runs easily, from what we have seen, one to two quarters,” Gupta said. Beyond installation sits the harder step of putting the system to work on live designs, and there “we are in single-digit customers that have moved to that second phase. That’s where we are. So still early days, I would say.”
It is an unusually frank framing in a market prone to declaring victory at launch. Siemens released the Fuse AI system, the generative layer that gives engineers natural-language access to tools, last year, followed by the agentic system early this year, after a period of working with what Gupta called “lighthouse” customers. Five months on from the agent’s debut, installation is broadening while production use remains thin.
NVIDIA lends the effort a marquee reference: the chipmaker publicly uses Siemens’ Fuse EDA solution in its own chip development, though it labels which products it runs in-house rather than claiming wall-to-wall adoption. For a technology this young, a named user of that stature matters, but it is a proof point, not a market.
The verification question, and the human in the loop
For chip designers, the sharpest worry about handing decisions to an agent is not speed but accountability. A wrong call in silicon costs months and millions. Gupta’s answer leans on the industry’s existing discipline rather than a new leap of faith.
He points to macro and memory placement, long treated as “part art, part science”, as work now largely ceded to automation, a shift he traces back to early machine-driven approaches such as Google’s AlphaChip. Engineers initially recoiled because machine-generated layouts did not look like human ones, even when they met every requirement. “Those days are gone,” he said. What has not changed is sign-off. Chip design still rests on four hard verification pillars–timing, physical verification, power integrity and test–each governed by “black and white” data-driven metrics. “The verification is very well defined, accountability is very clear,” Gupta said. An agent may explore the design space, but the sign-off gates it must clear remain the same.
That framework also shapes how Siemens answers the security question that governs whether agentic tools ever reach sensitive customers. Fuse runs on-premise, Gupta said, and an AI agent is subject to the same identity and access-management rules as a human engineer, with governance controls applied to the software Siemens ships. In an industry where taking design data off-site is, in his words, “a big no-no”, and where sovereign chip ambitions are sharpening that instinct, on-premise operation is less a feature than a precondition.
Open by necessity
The openness is real, but so is the gap between pitch and traction. Twenty-plus installations, single-digit customers in production, one to two quarters just to get set up: Fuse is a young system carrying a big strategic argument. Siemens is open because it has to be, and it is asking customers to treat that as a feature rather than a fallback.
Whether they do is the test. An orchestrator with no reason to lock you in is a genuinely different offer from one built to, but only if the tools underneath it deliver once the agent is doing the work. On that, Gupta is not overselling. “Still early days,” he said.
The post Why Siemens is selling open EDA while rivals build walled gardens appeared first on TechWire Asia.