“The line between AI workload and non-AI workload is getting blurrier by the day.” Prasad Kalyanaraman, VP of AWS Infrastructure Services, talks with Anurag Rana, Senior Technology Analyst at Bloomberg Intelligence, about what actually changes as AI scales. Prasad lays out the realities behind “AI data centers”: denser accelerators, non-blocking training networks, and modular in-chip liquid cooling. Underneath, the data-center basic priorities hold -— security as job zero, plus availability, performance, and cost. He also explains why clusters must be fungible between training and inference so accelerators and power never sit idle, and how AWS retrofits existing regions rather than treating AI as a separate build.