Steven Glinert and Mitch Nahmias talk with Gabriel Dickinson about using AI to help design analog and mixed-signal chips. They describe their company’s work on edge computing, including low-latency processing for physical systems.
The discussion examines Nvidia’s software ecosystem and interconnects, memory constraints and alternative approaches to inference. The guests describe possible competitive scenarios, while acknowledging that changing established hardware and software systems is difficult.
Steven Glinert and Mitch Nahmias offer different views on Chinese hardware and open models, the economics of training and inference, and possible co-design of models and chips. These are their analyses and forecasts rather than independently established outcomes.
Watch on YouTube
