Supporters of edge computing and on-device AI may be overly optimistic. The current key issue is that memory capacity and bandwidth have become the true bottlenecks of these architectures.

From a technical perspective, while offline AI models reduce network latency, they are limited by local device memory constraints, making large model deployment a serious challenge. In contrast, cloud computing, although involving network transmission, can access abundant memory resources, which still offers significant advantages in handling complex tasks.

Memory issues are not just about capacity but also involve access speed and bandwidth. If this infrastructure shortcoming is not overcome, the theoretical advantages of edge AI will be difficult to fully realize in practical applications.