Azure's 6.2× H100 markup vs RunPod

Same NVIDIA H100 SXM5, same 80 GB HBM3, same 700W TDP. Azure ND96isr-H100v5 on-demand: $12.29/h. RunPod H100 SXM5 on-demand: $1.99/h. Lambda Labs sits in between at $2.99/h. The hardware doesn't change. The bill does.

Azure's three-year reservation drops the rate to $5.47/h — still 2.7× the on-demand RunPod price. The premium pays for first-party support, integrated networking, and the ability to expense it on the enterprise contract — but the GPU itself is identical to what an independent hoster rents you for one-sixth the price.

The TokenFloor GPU TCO calculator surfaces this directly: at 80% utilization, a single owned H100 (capex $35k, 3-year horizon, EU electricity at $0.30/kWh, PUE 1.4) costs around $1500/month all-in (depreciation + power + rack + sysadmin). The same workload on Azure on-demand: $7180/month. On RunPod: $1162/month. Pick your provider before you pick your model — the GPU-cloud markup is often larger than the model-vendor markup.