A presentation on AI infrastructure with a focus on large-scale, distributed training and inference. In this presentation, we aim to introduce a new framework for generative AI-optimized infrastructure, drawing on traditional HPC-based approaches. Building infrastructure for artificial intelligence requires a system designed to handle extremely demanding computational tasks. These systems must support massive parallel processing, high-speed communication, and efficient data management to train and deploy large models effectively. Unlike traditional workloads, AI operations depend on seamless collaboration between computing, storage, and networking resources. To meet these needs, the infrastructure must be highly scalable, support distributed workloads, and ensure low latency. The overall goal is to create an environment where complex models can be trained and used efficiently, without becoming a bottleneck to progress.