Loading benchmark data
The H200 is an H100 with 141GB of faster HBM3e memory, which lifts throughput on models too large to sit comfortably in 80GB. It reached the clouds through 2024 and 2025, so its price series is short by construction rather than by omission.
Global median across 3 providers and 48 regions, Aug 2026.
Azure entered the benchmark this month, so this figure is not directly comparable with the month before it.
Jan 2025: Google Cloud enters the benchmark. A step here is a change in what is being measured, not a price move.
Aug 2026: Azure enters the benchmark. A step here is a change in what is being measured, not a price move.
Azure publishes no price history, so its series begins in Aug 2026, when GPUQuant recorded its first snapshot. It cannot be backfilled.
Prices are Linux on-demand list prices for the H200 141GB SXM5, eight-GPU node, normalized to one GPU-hour. How the benchmark works
Newest month GPUQuant holds. Each provider figure is the median of its own regional prices, so a provider that publishes more regions does not count for more.
| Provider | Regional median | GPU-month | Lowest region | Highest region | Regions | History begins |
|---|---|---|---|---|---|---|
| AWS | $9.49 | $6,931 | $7.91us-east-1 | $10.92ap-southeast-3 | 11 | Dec 2024? |
| Azure | $13.25 | $9,673 | $10.60eastus2 | $21.20brazilsouth | 31 | Aug 2026? |
| Google Cloud | $10.84 | $7,917 | $10.84us-east4 | $12.77asia-south1 | 6 | Jan 2025? |
One dot per region, at the lowest eligible price published there. The bar spans that provider's cheapest to dearest region.
| AWS | us-east-2 | $7.9120 |
| AWS | us-west-2 | $7.9120 |
| AWS | us-east-1 | $7.9120 |
| AWS | eu-north-1 | $8.4658 |
| AWS | eu-south-2 | $8.5450 |
| AWS | me-central-1 | $9.4944 |
| AWS | ap-south-1 | $9.4944 |
| AWS | ap-northeast-1 | $9.8900 |
| AWS | us-west-1 | $9.8900 |
| Azure | westus2 | $10.6000 |
| Azure | eastus2 | $10.6000 |
| Azure | westus3 | $10.6000 |
| Google Cloud | us-east4 | $10.8450 |
| Google Cloud | us-west1 | $10.8450 |
| Google Cloud | us-central1 | $10.8450 |
| Google Cloud | us-east1 | $10.8450 |
| AWS | ap-southeast-3 | $10.9186 |
| AWS | ap-northeast-2 | $10.9186 |
| Azure | southcentralus2 | $11.6600 |
| Azure | mexicocentral | $11.6600 |
| Google Cloud | europe-west1 | $11.9414 |
| Azure | centralus | $11.9780 |
| Azure | indonesiacentral | $12.4020 |
| Azure | malaysiawest | $12.4020 |
| Azure | northeurope | $12.7200 |
| Azure | northcentralus | $12.7200 |
| Google Cloud | asia-south1 | $12.7687 |
| Azure | usgovvirginia | $13.2500 |
| Azure | uksouth | $13.2500 |
| Azure | usgovarizona | $13.2500 |
| Azure | koreasouth | $13.2500 |
| Azure | japaneast | $13.2500 |
| Azure | usgovarizona | $13.2500 |
| Azure | francecentral | $13.2500 |
| Azure | westus | $13.7800 |
| Azure | southeastasia | $13.7800 |
| Azure | spaincentral | $13.7800 |
| Azure | germanywestcentral | $13.7800 |
| Azure | italynorth | $13.7800 |
| Azure | australiasoutheast | $13.7800 |
| Azure | swedencentral | $13.7840 |
| Azure | ukwest | $13.8860 |
| Azure | norwayeast | $15.1580 |
| Azure | uaenorth | $15.1580 |
| Azure | australiaeast | $15.3700 |
| Azure | southafricanorth | $15.4760 |
| Azure | polandcentral | $16.3240 |
| Azure | japanwest | $16.4300 |
| Azure | brazilsouth | $21.2000 |
49 eligible offerings across 48 regions. The benchmark uses the lowest eligible price in each region, so where a provider lists several equivalent nodes only the cheapest enters the median.
Azure history begins Aug 2026, when GPUQuant started recording snapshots. Azure publishes no price archive, so nothing earlier can ever be recovered.
AWS p5en and p5e, Azure ND H200 v5 and Google Cloud a3-ultragpu-8g.
A PCIe variant none of the three providers publishes an on-demand list price for.
Held constant across providers for the same reason as H100.
Every figure on this page is one GPU for one hour. Providers do not publish that number, so GPUQuant derives it.
AWS prices the whole instance. GPUQuant divides the Linux on-demand hourly rate by the GPU count in the AWS price file, which matches the count in the AWS instance specification tables.
Azure prices the whole virtual machine and its size name does not encode a GPU count, so GPUQuant divides by a count read from the Microsoft size-series documentation.
Google prices accelerators, vCPU and memory as separate SKUs, so GPUQuant rebuilds the complete machine price from its components before dividing by GPU count. Anything less would compare a bare accelerator against two whole machines.
Sources: AWS EC2 Price List, Azure Retail Prices API, Google Cloud Billing Catalog. GPU counts are read from the provider specification tables named on each offering row. See methodology for the full definition. Providers currently publishing: Amazon Web Services, Microsoft Azure, Google Cloud.