AI agents turn one request into a chain of inference calls, and that's what blows up the budget. Vasanth Mohan of
@SambaNovaAI broke down tokens per second per chip, model tiering, and how their RDU and disaggregated inference keep agent costs in check.