Cadence and NVIDIA Build Industry's First Silicon-to-System Agentic AI Engineering Platform

For decades, the semiconductor industry has dealt with rising design complexity in the same way: adding more engineers, more computing power, and more time. That worked when complexity grew gradually, but it no longer does. Design teams that once numbered in the dozens now run into the thousands, verification alone consumes billions of computing hours each year, and tasks that used to take days can now stretch into weeks. At DAC 2026, Cadence and NVIDIA are proposing a different approach: autonomous AI agents capable of understanding what a design is meant to do, reasoning through engineering problems, and carrying out significant parts of the design process with minimal human oversight.

The shift Cadence describes is a step beyond generative AI that simply helps engineers write code, toward AI systems that can own entire engineering tasks, evaluate their own results, and keep working toward a finished outcome. Central to this is NVIDIA's Nemotron 3 Ultra model, which currently leads among open models on a benchmark for agentic RTL coding, meaning AI-driven generation of the hardware description code used to define chip logic. Paired with NVIDIA Research's ACE-RTL agent, it goes beyond generating code to reasoning about hardware architecture and iterating on its own work to reach functionally correct designs. Enterprises can also customize and deploy the model using t...