What are AI servers mainly used for, and how do training and inference machines differ?

They serve large-model training, fine-tuning, vector retrieval, AI inference and compute simulation. Training machines: multi-GPU with high-speed interconnect — mainly H100/B300, large memory, strong compute. Inference machines: fewer GPUs, such as B10/A10, prioritizing low latency and high throughput at lower cost.