GPU Economics of Self-Hosting Open Models: H200 Loses Money Even Fully Loaded, B300 Profits at 35% Utilization

An analysis by Yingjun Wu, amplified by SemiAnalysis, finds that GPU choice matters far more than inference engine choice for hosting open models: H200 loses money even at full utilization, while B300 turns profitable at 35% utilization.

2026-11-03 ~ 2026-11-03 · 2 related posts

• H200 loses money even fully busy: GPU choice beats inference engine for hosting economics — teortaxesTex · 2026-11-03
• H200 loses money at full utilization while B300 breaks even at 35%: GPU hosting economics for open-source models — AccBalanced · 2026-11-03