AI Inference Silicon
Inference silicon runs trained models in production rather than training them, and rewards throughput per watt over raw peak performance. It is a distinct design problem from training, and has drawn in custom accelerator designers alongside the incumbent chip vendors.
7 stocks
