Cadence Design Systems Leverages Small Language Models for Edge AI Innovation in Q2 2025

The topic of "Small language models" appears in Cadence Design Systems, Inc.'s Q2 2025 earnings transcript primarily in the context of physical AI and edge device applications. The discussion occurs during the Q&A segment, with an analyst asking about the impact of developments around edge devices and small language models on Cadence’s bookings and product demand.

Context and Company Perspective

• The analyst frames the question by noting that key development partners have recently highlighted their work with edge devices and small language models as enablers of physical AI.
• Cadence’s CEO, Anirudh Devgan, responds by emphasizing broad-based demand driven by AI innovation, including physical AI, which he sees as a significant growth area alongside AI infrastructure and sciences AI.
• He explains that physical AI involves silicon chips optimized for power and battery efficiency in edge devices like autonomous cars, robots, and drones, which differ from data center silicon.
• Small language models are referenced as part of the AI models used in physical AI, described as "more word model than LLMs (large language models)."
• Devgan highlights that while inference runs on edge devices, training still occurs in data centers, creating a dual opportunity for Cadence’s products in both physical AI silicon design and AI infrastructure.

Strategic Implications

• Cadence views physical AI, including applications involving small language models, as a key driver of future growth.
• The company’s privileged position working with top semiconductor customers allows it to participate early in R&D for these emerging AI applications.
• The differentiation between AI silicon for edge devices (power-optimized, battery-efficient) and data center silicon suggests Cadence’s tools must support diverse design requirements.
• The mention of small language models in this context signals Cadence’s recognition of evolving AI workloads beyond large-scale data center models, expanding into embedded, real-time AI applications.
• This evolution is contributing to bookings strength and increased customer investment in Cadence’s design tools.

Relevant Quote

"The silicon is different in the car or in the robot or in the edge devices... They are more word model than LLMs. But all these physical AIs still need to be trained. Even if the inference like for autonomous car runs on the car, the actual AI model is trained on the data center. So the beautiful thing of physical AI is not only it creates new opportunities for us, it also emphasizes the importance of AI infrastructure in the data centers. So it is helping both sides of that equation. And so we are benefiting from that."

Summary

Cadence Design Systems positions small language models as an integral part of the physical AI wave, particularly in edge computing devices requiring specialized silicon design. This trend is driving increased R&D investment from customers and strengthening Cadence’s bookings. The company’s strategic focus on supporting both edge AI silicon and data center AI infrastructure positions it well to capitalize on the growing adoption of small language models in real-world AI applications.