The H100 GPU’s hourly price varies dramatically, ranging from $2.69 to $12.29 per GPU‑hour depending on the provider. The cheapest option is Run Pod Community at $2.69, which totals $1,964 for a month of continuous use, while the most expensive listed is Azure ND96isr_H100_v5 at $12.29 per hour, amounting to $8,972 monthly. These figures are grouped by exact model and memory configuration because “H100” actually covers three distinct products: the SXM 80 GB, the PCIe 80 GB and the NVL 94 GB.
By keeping comparisons inside a single (model, memory) group the spread reflects real hardware differences rather than artificial chart noise. Within a vendor, Azure shows the same A100 PCIe 80 GB priced differently based on VM size, but the spread uses each vendor’s cheapest route to a given GPU. The 4.57× gap is not pure margin.
Azure’s ND series provides 400 Gbps InfiniBand with GPU‑Direct RDMA, essential for multi‑node training, which justifies the higher rate versus a basic single‑GPU SKU. Run Pod Community offers third‑party hardware without enterprise‑grade SLAs or committed capacity, so users fine‑tuning on a single node pay for an interconnect they may never use. The analysis is based on on‑demand list prices from AWS us‑east‑1 and Azure eastus, excluding Lambda and spot pricing.
Source: Hacker News · Summarized by HeadlinesBriefing