The post analyzes the unit economics of AI infrastructure, specifically comparing the H100 and B200 GPUs. Calculations suggest that an H100 pod can generate approximately $158/hr in profit with a 59% gross margin, while a B200 pod could reach $405/hr in profit with an 82% gross margin. The analysis notes that profitability is highly dependent on batching and concurrent users, as costs remain fixed regardless of utilization until the KV cache is full.