CDNS: Mgmt Meetings Highlight Strong Chip Design Activity, Confidence in Agentic AI Positioning, & More

Research evidence excerpt

CDNS: Mgmt Meetings Highlight Strong Chip Design Activity, Confidence in Agentic AI Positioning, & More

ased for capacity and through its token and card models. Our discussions noted Cadence’s focus

on value-based pricing for its new agentic AI offerings vs. discounting to drive faster adoption. 2)

Middle Layer = Core EDA. Cadence’s core EDA subscription model remains the anchor arrangement

with customers. Agentic AI does not replace Cadence’s core EDA engines, but rather calls them more

often and in an intelligent manner resulting in additional exploration, verification, optimization and

compute. 3) Base Layer = Compute / Data.

Discussions highlighted Cadence's AI portfolio as including:

Optimization AI. Cadence noted that its optimization AI is built on purpose-specific neural networks

that are tightly integrated into its EDA tools. These are lightweight, domain-trained language models

that are optimized for real-time reinforcement learning during active design execution. From a product

standpoint, we'd highlight:

• Cerebrus. In digital implementation the company's Cerebrus offering applies reinforcement

learning to explore backend design spaces autonomously - capable of large-scale compute

allocation across 100s to 1,000s of CPUs for design exploration and capable of cross project

learning where prior designs exist.

• Verisium. With verification continuing to represent one of longest and most compute-intensive

stages of the design flow, traditional stimulus generation has been largely random and inefficient.

However, Cadence noted Verisium's intro of reinforcement learning as capable of correlating

stimulus behavior with design state coverage - capable of reducing regression cycle time from 24