Pomegra
Agentrys secured $24.5M across a pre-seed and an oversubscribed seed round to deploy AI agents that carry semiconductor designs from specification to sign-off-clean GDS layout without human engineers.
Key Takeaways:
- A $19.1M seed led by Etna Labs and a prior $5.4M pre-seed backed by MediaTek bring Agentrys' total raise to $24.5M.
- The platform already ran a 32-bit CPU from specification to sign-off-clean GDS layout autonomously, with no human in the loop.
- Agentrys is among the first to exceed 90% accuracy on NVIDIA's public CVDP verification benchmark.
Lead
San Jose-based Agentrys announced $24.5 million in total funding on August 26, 2026 - an oversubscribed $19.1 million seed round led by Etna Labs, following a $5.4 million pre-seed that MediaTek led as the company's first strategic backer. Valuation was not disclosed. The startup is building what it calls Agentic Design Automation (ADA), software that deploys fleets of AI agents across semiconductor design workflows, handling the full pipeline from an initial specification to a sign-off-clean GDS layout file with no human engineer touching it.
What Does Agentrys Actually Build?
ADA moves past today's Electronic Design Automation tools, which automate individual design steps but still require engineers to stitch them together, catch failures, and manage handoffs. Agentrys replaces that orchestration with autonomous agents that execute tasks, monitor results, self-correct, and iterate until a design clears sign-off. The target is the engineering hours that accumulate between a written specification and a file ready for a foundry.
The company demonstrated its platform end to end. Its agents autonomously took a 32-bit CPU from specification to a sign-off-clean GDS layout, the final format submitted to a fab for fabrication. On NVIDIA's publicly available CVDP verification benchmark, Agentrys is among the first vendors to exceed 90% accuracy. An internal agent evolution technique improves performance across successive iterations rather than treating each run as independent.
Who Is Behind Agentrys?
Mark Ren founded the company on nearly three decades of EDA and AI research at NVIDIA Research and IBM Research. He led ChipNeMo, the first industrial large language model trained specifically for chip design tasks. That project put NVIDIA in an early position in AI-native EDA and gave Ren a detailed map of where conventional tools stall. The founding thesis is built directly on that map.
Why Did the Seed Oversubscribe?
Chip design has not scaled in proportion to chip complexity. The engineering headcount required to tape out a leading-edge SoC has grown faster than the supply of qualified engineers, and existing EDA vendors have not closed that gap with software alone. An oversubscribed seed round suggests multiple institutional investors converged on the same opening before pricing settled.
MediaTek writing the pre-seed check is the more telling data point. Strategic investors at that stage are rare - they tend to wait until a product is proven enough to defend the check internally. Its participation implies the platform was delivering something useful against real tapeout problems, not just benchmark results. Current customers include major fabless semiconductor companies, a global foundry, and early-stage chip startups, spanning both digital and analogue design flows.
What the Round Implies About the Previous One
The jump from $5.4 million to $19.1 million in a single funding cycle is steep. That trajectory is consistent with a company that moved from research-grade demonstration to active customer use across multiple design flow types in a compressed period. It also sets a high bar for the Series A. The platform will need to show cycle-time reductions on production designs, not just on a 32-bit CPU chosen to showcase the workflow. Larger designs carry more constraints, tighter timing budgets, and less tolerance for autonomous agents making suboptimal routing decisions.
New capital will fund recruitment, agent-native tooling, and expanded customer engagements across verification and physical design.
Outlook
Agentrys enters a narrow but increasingly competitive space, as chipmakers and venture funds alike back agentic EDA startups. Its near-term advantage is a founder with direct industrial LLM experience in chip design, an end-to-end demonstrated flow on a real processor, and two strategic customers willing to share early data. Whether autonomous GDS generation generalizes from a 32-bit CPU to the larger, more constrained designs at the center of commercial production is the open question. The Series A will likely hinge on that answer.
