Part 3/10:
During an exclusive interview on the All-in Summit, Musk provided insights into Tesla’s chip development strategy. Historically, Tesla operated two key chip programs: one for training AI models (Dojo) and the other for inference (AI4). The current inference chip, AI4, is already in operational use across Tesla vehicles, enabling advanced driver-assistance features.
Tesla is now finalizing the design of AI5, which promises a 40-fold improvement over AI4 by some metrics, and an eightfold increase in raw computational power. This leap enables better handling of AI operations such as softmax calculations, which are integral to neural network inference, reducing latency significantly—an order-of-magnitude improvement in performance.