The HPC community built the world's fastest computers by understanding something that the AI industry is only now discovering: the network is not a support layer — it is the performance multiplier that determines whether your compute investment pays off. This talk introduces Networkonomics, a framework for thinking about high-performance networking as an economic decision rather than just a technical one. We examine how network properties — lossless delivery, sub-microsecond latency, high message rate, advanced routing — translate directly into measurable outcomes: simulation throughput, training efficiency, job completion assurance, and inference revenue at scale. As HPC clusters converge with AI factories, and as agentic workloads add new demands for high-rate small-message traffic, the network's role in system economics only grows. We'll look at where high-performance networking is headed, why the lessons of exascale HPC are the right foundation for AI infrastructure design, and what a fabric-first economic framework means for the infrastructure decisions today and through the end of the decade.
