HN748: How AI and HPC Are Changing Data Center Networks

This podcast episode dives into the critical advancements and challenges within AI training infrastructure, underscoring the necessity of dedicated networks to manage bursty workloads, the evolution of NICs designed for efficient GPU interactions, and addressing the pressing power and cooling limitations that arise as bandwidth demands escalate. The discussion highlights innovative solutions such as optical circuit switching and liquid cooling, while forecasting significant opportunities and disruptions in the AI infrastructure space, ultimately suggesting a robust future for AI technologies beyond the current hype.