Agentrys, a startup developing artificial intelligence systems for semiconductor engineering, has raised $24.5 million as chipmakers look to automate increasingly complex design and verification workflows.
The funding consists of a $19.1 million seed round led by Etna Labs and a $5.4 million pre-seed round led by MediaTek (TPE: 2454), which became Agentrys’ first strategic investor. The company said the latest financing will support hiring, development of agent-native tools and expansion of customer engagements across verification and physical design.
Agentrys is targeting a semiconductor industry where design complexity has increased alongside demand for advanced processors for artificial intelligence, high-performance computing, automotive systems and other applications. Despite decades of progress in electronic design automation (EDA), many chip-development workflows continue to require specialized engineers and substantial manual effort.
The company was founded by Mark Ren, who has nearly three decades of experience in EDA and AI research, including work at NVIDIA (NASDAQ: NVDA) Research and IBM (NYSE: IBM) Research. Ren also led ChipNeMo, an industrial large language model initiative focused on semiconductor design.
Agentrys is building what it calls Agentic Design Automation, or ADA. The approach extends conventional EDA beyond individual automated tasks by using AI agents to coordinate broader engineering workflows, learn from results and improve their performance over time.
Traditional EDA platforms automate specific portions of the design process, such as simulation, synthesis, verification and physical implementation. Agentrys aims to place AI agents across these workflows, allowing engineering teams to automate sequences of tasks rather than treating each activity as a separate operation.
The company says its platform is designed as an open system that customers can customize. Instead of relying exclusively on predefined agents controlled by a software vendor, semiconductor companies can build and manage their own AI-based engineering workforce around their existing commercial EDA tools, internal software and infrastructure.
A second component is Agentrys’ design intelligence layer, which the company says can learn from customer data, system usage and evaluation results. Its agent-native tools and specialized models are intended to extend the capabilities of general-purpose AI systems into semiconductor engineering.
That learning component is central to Agentrys’ longer-term strategy. The company wants engineering workflows to improve through repeated execution, with each task generating data that can be evaluated and used to refine subsequent performance.
“Building production-grade agents that reliably automate real engineering work is far from easy — that’s the problem Agentrys exists to solve,” Ren, founder and CEO of Agentrys, said in a statement.
The company has already demonstrated an autonomous multi-agent workflow that took a 32-bit CPU from specification through sign-off-clean GDS layout without human intervention, according to Agentrys.
Verification is another area where the company claims early results. On NVIDIA’s public CVDP verification benchmark, Agentrys said it has become one of the first systems to exceed 90% accuracy. The company said its agent-evolution techniques have improved performance through successive iterations.
The benchmark results are relevant because verification is among the most time-consuming stages of chip development. As designs become larger and more sophisticated, engineers must identify functional errors and validate increasingly complex interactions before a design can move into manufacturing.
Agentrys is positioning its technology around this wider engineering bottleneck. The company said it is already working with several major fabless semiconductor companies, a global foundry and chip startups. Its current engagements cover digital and analog design flows, while system design is on the roadmap.
Investors are betting that semiconductor engineering could become an important application for agentic AI because the output of many design workflows can be measured against objective engineering criteria.
Etna Labs said chip design provides an environment in which AI systems can repeatedly execute tasks and evaluate results using established engineering tools and metrics. That creates a potential pathway for systems that improve through repeated design cycles.
MediaTek Innovation Fund, meanwhile, said semiconductor research and development presents domain-specific challenges for agentic AI. Brian Hsu, managing director of the fund, said Agentrys’ approach could allow engineering teams to customize agents and capture domain knowledge for reuse across semiconductor R&D.
The opportunity comes as semiconductor companies face pressure to shorten development cycles while managing the rising cost and complexity of advanced chip design. The global semiconductor industry, which Agentrys estimates at more than $800 billion, increasingly depends on specialized engineering talent and sophisticated software tools.
For chipmakers, the potential value of agentic design systems will depend on whether they can operate reliably within production environments and integrate with existing EDA infrastructure. Semiconductor design errors can carry significant financial and schedule consequences, making accuracy and predictability essential.
Agentrys said its objective is to build an engineering workforce that becomes more capable through continued use. The new funding will allow the company to expand that approach across additional stages of the design process and make its platform available to more semiconductor engineering teams.
The financing also places Agentrys within a growing group of startups attempting to apply AI agents to highly specialized technical workflows. In semiconductor design, however, the test will be whether autonomous systems can move beyond demonstrations and consistently deliver results that meet the stringent requirements of commercial chip development.
