Decode waits on memory, so we priced H100, H200 and B200 per TB/s-hour from NVIDIA's specs and 2026-10-30 list prices. M...

Decode waits on memory, so we priced H100, H200 and B200 per TB/s-hour from NVIDIA's specs and 2026-10-30 list prices. Modal's H200 costs 15% more than its H100 for 43% more bandwidth. Break-even is 1.43x, and only Together AI's list price misses it. When you pick a GPU for serving, do you compare bandwidth, memory size or the hourly price?https://inferprice.pages.dev/articles/h200-h100-cost-per-tb-s#AI #LLM #Python #GPU #FinOps