Cadence Design Systems on Monday introduced Maestro, a set of specialized AI models that power its chip-design agents, built in part on Nvidia's open Nemotron models. The San Jose company announced it hours before reporting quarterly earnings after the bell.
Why it matters: The two giants of chip-design software are now taking different paths on AI. Cadence lets customers run and retrain the models on their own machines, while rival Synopsys built its model with OpenAI on OpenAI-hosted infrastructure.
What Cadence claims
Cadence says early tests show up to 30% higher quality results and up to five times better token efficiency than the general-purpose model Maestro started from.
The first models go to select early-access customers in the fourth quarter. Customers can run them on Cadence's cloud, public clouds or their own on-premises systems, and can post-train them on their own designs and data.
Why run it in-house
Paul Cunningham, a Cadence senior vice president, told Tom's Hardware that agent costs have "really started to explode," and that moving more work into the model weights cuts that cost. "You don't need to drive to Safeway in the jet engine," he said.
Cadence said it is also working with Nvidia on security using Nvidia's OpenShell reference architecture. Reuters reported the move is aimed at chipmakers wary of sending sensitive designs to outside AI providers.