Loading benchmark data
The H100 is the GPU most large language models were trained on between 2023 and 2025, and it remains the reference unit for AI compute pricing. It is one of the two GPUs CME Group and Silicon Data plan to list rental-index futures on.
Global average across 4 providers and 123 regions, Sep 2026.
Feb 2024: Google Cloud enters the benchmark. A step here is a change in what is being measured, not a price move.
Aug 2026: Oracle 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 Dec 2023, when GPUQuant recorded its first snapshot. It cannot be backfilled.
Oracle 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 H100 80GB SXM5, eight-GPU node, normalized to one GPU-hour. How the benchmark works
Newest month GPUQuant holds. Each provider figure is the average of its own regional prices, so a provider that publishes more regions does not count for more. The median is shown beside it: the two rarely differ by much, and where they do it is because the GPU is sold in only a handful of regions.
| Provider | Regional average | GPU-month | Median | Lowest region | Highest region | Regions | History begins |
|---|---|---|---|---|---|---|---|
| AWS | $8.45 | $6,166 | $8.60 | $6.88us-east-1 | $11.56sa-east-1 | 13 | Jul 2023 |
| Azure | $14.77 | $10,783 | $14.38 | $11.06eastus | $23.33southafricawest | 24 | Dec 2023 |
| Google Cloud | $13.41 | $9,789 | $13.19 | $10.98us-central1 | $17.57me-central2 | 41 | Feb 2024 |
| Oracle | $10.00 | $7,300 | $10.00 | $10.00af-casablanca-1 | $10.00us-sanjose-1 | 45 | Aug 2026 |
One dot per region, at the lowest eligible price published there. The bar spans that provider's cheapest to dearest region.
Every region a provider publishes this GPU in, at the price that enters the benchmark.
| Sizes | |||
|---|---|---|---|
| AWS | Ohious-east-2Columbus | $6.8800 | |
| AWS | Northern Virginiaus-east-1 | $6.8800 | |
| AWS | Oregonus-west-2 | $6.8800 | |
| AWS | Stockholmeu-north-1 | $7.3616 | |
| AWS | Montrealca-central-1 | $7.9120 | |
| AWS | Mumbaiap-south-1 | $8.2560 | |
| AWS | Northern Californiaus-west-1San Jose | $8.6000 | |
| AWS | Tokyoap-northeast-1 | $8.6000 | |
| AWS | Sydneyap-southeast-2 | $8.9440 | |
| AWS | Londoneu-west-2 | $8.9440 | |
| AWS | Seoulap-northeast-2 | $9.4944 | |
| AWS | Jakartaap-southeast-3 | $9.4944 | |
| Oracle | Sydneyap-sydney-1 | $10.0000 | |
| Oracle | Melbourneap-melbourne-1 | $10.0000 | |
| Oracle | Sao Paulosa-saopaulo-1São Paulo | $10.0000 | |
| Oracle | Vinhedosa-vinhedo-1 | $10.0000 | |
| Oracle | Montrealca-montreal-1 | $10.0000 | |
| Oracle | Torontoca-toronto-1 | $10.0000 | |
| Oracle | Santiagosa-santiago-1 | $10.0000 | |
| Oracle | Valparaisosa-valparaiso-1 | $10.0000 | |
| Oracle | Bogotasa-bogota-1 | $10.0000 | |
| Oracle | Pariseu-paris-1 | $10.0000 | |
| Oracle | Marseilleeu-marseille-1 | $10.0000 | |
| Oracle | Frankfurteu-frankfurt-1 | $10.0000 | |
| Oracle | Hyderabadap-hyderabad-1 | $10.0000 | |
| Oracle | Mumbaiap-mumbai-1 | $10.0000 | |
| Oracle | Batamap-batam-1 | $10.0000 | |
| Oracle | Jerusalemil-jerusalem-1 | $10.0000 | |
| Oracle | Milaneu-milan-1 | $10.0000 | |
| Oracle | Turineu-turin-1 | $10.0000 | |
| Oracle | Osakaap-osaka-1 | $10.0000 | |
| Oracle | Tokyoap-tokyo-1 | $10.0000 | |
| Oracle | Kulaiap-kulai-2 | $10.0000 | |
| Oracle | Queretaromx-queretaro-1 | $10.0000 | |
| Oracle | Monterreymx-monterrey-1 | $10.0000 | |
| Oracle | Casablancaaf-casablanca-1 | $10.0000 | |
| Oracle | Amsterdameu-amsterdam-1 | $10.0000 | |
| Oracle | Riyadhme-riyadh-1 | $10.0000 | |
| Oracle | Jeddahme-jeddah-1 | $10.0000 | |
| Oracle | Jovanovaceu-jovanovac-1 | $10.0000 | |
| Oracle | Singaporeap-singapore-1 | $10.0000 | |
| Oracle | Singaporeap-singapore-2 | $10.0000 | |
| Oracle | Johannesburgaf-johannesburg-1 | $10.0000 | |
| Oracle | Seoulap-seoul-1 | $10.0000 | |
| Oracle | Chuncheonap-chuncheon-1 | $10.0000 | |
| Oracle | Madrideu-madrid-1 | $10.0000 | |
| Oracle | Madrideu-madrid-3 | $10.0000 | |
| Oracle | Stockholmeu-stockholm-1 | $10.0000 | |
| Oracle | Zuricheu-zurich-1 | $10.0000 | |
| Oracle | Abu Dhabime-abudhabi-1 | $10.0000 | |
| Oracle | Dubaime-dubai-1 | $10.0000 | |
| Oracle | Londonuk-london-1 | $10.0000 | |
| Oracle | Newportuk-cardiff-1 | $10.0000 | |
| Oracle | Northern Virginiaus-ashburn-1 | $10.0000 | |
| Oracle | Chicagous-chicago-1 | $10.0000 | |
| Oracle | Phoenixus-phoenix-1 | $10.0000 | |
| Oracle | San Joseus-sanjose-1 | $10.0000 | |
| Google Cloud | Iowaus-central1 | $10.9791 | |
| Google Cloud | South Carolinaus-east1 | $10.9791 | |
| Google Cloud | Oregonus-west1 | $10.9791 | |
| Google Cloud | Northern Virginiaus-east4 | $10.9791 | |
| Google Cloud | Phoenixus-west8 | $10.9791 | |
| Azure | Virginiaeastus | 5 sizes | $11.0610 |
| Azure | Phoenixwestus3 | 4 sizes | $11.0610 |
| Azure | Virginiaeastus2 | 6 sizes | $11.0610 |
| Azure | Washingtonwestus2 | 4 sizes | $11.0610 |
| Google Cloud | Mumbaiasia-south1 | $11.4182 | |
| AWS | São Paulosa-east-1 | $11.5584 | |
| Google Cloud | Tel Avivme-west1 | $12.0770 | |
| Google Cloud | Haminaeurope-north1 | $12.0890 | |
| Google Cloud | Belgiumeurope-west1 | $12.0890 | |
| Google Cloud | Montrealnorthamerica-northeast1 | $12.0890 | |
| Google Cloud | Torontonorthamerica-northeast2 | $12.0890 | |
| Google Cloud | Columbusus-east5 | $12.3657 | |
| Google Cloud | Las Vegasus-west4 | $12.3657 | |
| Google Cloud | Londoneurope-west2 | $12.5161 | |
| Google Cloud | Taiwanasia-east1 | $12.7127 | |
| Google Cloud | Milaneurope-west8 | $12.7357 | |
| Google Cloud | Pariseurope-west9 | $12.7357 | |
| Google Cloud | Frankfurteurope-west3 | $12.9553 | |
| Google Cloud | Madrideurope-southwest1 | $12.9553 | |
| Google Cloud | Dallasus-south1 | $12.9553 | |
| Google Cloud | Delhiasia-south2 | $13.1869 | |
| Google Cloud | Los Angelesus-west2 | $13.1979 | |
| Google Cloud | Salt Lake Cityus-west3 | $13.1979 | |
| Azure | Texassouthcentralus | 4 sizes | $13.2714 |
| Azure | Illinoisnorthcentralus | 5 sizes | $13.2714 |
| Azure | Wyomingwestcentralus | 3 sizes | $13.2733 |
| Google Cloud | Dohame-central1 | $13.3396 | |
| Azure | Iowacentralus | 5 sizes | $13.6050 |
| Google Cloud | Sydneyaustralia-southeast1 | $13.7238 | |
| Azure | Londonuksouth | 2 sizes | $13.8262 |
| Google Cloud | Netherlandseurope-west4 | $13.9771 | |
| Google Cloud | Osakaasia-northeast2 | $14.0993 | |
| Google Cloud | Seoulasia-northeast3 | $14.0993 | |
| Google Cloud | Warsaweurope-central2 | $14.1454 | |
| Google Cloud | Turineurope-west12 | $14.1630 | |
| Google Cloud | Singaporeasia-southeast1 | $14.1944 | |
| Google Cloud | Melbourneaustralia-southeast2 | $14.2728 | |
| Google Cloud | Johannesburgafrica-south1 | $14.3606 | |
| Azure | Warsawpolandcentral | 5 sizes | $14.3793 |
| Azure | Californiawestus | 4 sizes | $14.3793 |
| Azure | United Statessoutheastus | 4 sizes | $14.3793 |
| Azure | Madridspaincentral | 5 sizes | $14.3793 |
| Azure | Netherlandswesteurope | 4 sizes | $14.3793 |
| Azure | Gävleswedencentral | 6 sizes | $14.3793 |
| Azure | Torontocanadacentral | $14.7480 | |
| Google Cloud | Jakartaasia-southeast2 | $14.7635 | |
| Azure | Seoulkoreacentral | 2 sizes | $14.9324 |
| Google Cloud | Zuricheurope-west6 | $15.3586 | |
| Google Cloud | Hong Kongasia-east2 | $15.3586 | |
| Google Cloud | Santiagosouthamerica-west1 | $15.7001 | |
| Google Cloud | Tokyoasia-northeast1 | $15.7236 | |
| Azure | Dubaiuaenorth | 2 sizes | $15.8172 |
| Azure | Zurichswitzerlandnorth | 2 sizes | $15.8172 |
| Azure | Tokyojapaneast | 2 sizes | $16.0385 |
| Google Cloud | Berlineurope-west10 | $16.9078 | |
| Azure | Osakajapanwest | 2 sizes | $17.1445 |
| Google Cloud | São Paulosouthamerica-east1 | $17.4336 | |
| Google Cloud | Dammamme-central2 | $17.5665 | |
| Azure | New South Walesaustraliaeast | $17.8250 | |
| Azure | Johannesburgsouthafricanorth | $21.0840 | |
| Azure | Cape Townsouthafricawest | $23.3260 |
123 regions across 73 places, 179 eligible offerings. The benchmark uses the lowest eligible price in each region, so where a provider lists several equivalent sizes only the cheapest enters the average.
Azure history begins Dec 2023, when GPUQuant started recording snapshots. Azure publishes no price archive, so nothing earlier can ever be recovered.
Azure history begins Aug 2026, when GPUQuant started recording snapshots. Azure publishes no price archive, so nothing earlier can ever be recovered.
AWS p5, Azure ND H100 v5 and Google Cloud a3-highgpu-8g. The same silicon in the same NVLink-connected form factor, which is what makes a cross-provider benchmark meaningful.
A PCIe card with 94GB and no NVLink fabric across eight GPUs. Different capacity and different interconnect, so it is a different product rather than a cheaper H100.
Google Cloud prices these above standard H100. They are the same 80GB silicon on a higher networking tier, and including them would give Google Cloud two quotes in the same region.
A smaller shape carries a different share of host CPU and memory per GPU. Holding node size constant is what keeps the three providers comparable.
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.
Oracle already prices per GPU-hour and bills its bare-metal GPU shapes as a whole shape, with no separate OCPU or memory SKU for any GPU family. The published figure is therefore already a complete per-GPU number and GPUQuant does not divide it.
Frontier-scale training and low-latency inference on very large models.
Memory-bound inference and training of larger models.
Established training and inference workloads, and price-sensitive capacity.
H100 is one of the two GPUs in the planned CME Group and Silicon Data rental-index futures. What that means. GPUQuant's reference price is an independently calculated list-price benchmark and is not the settlement index for Silicon Data H100 Rental Index Futures.
Sources: AWS EC2 Price List, Azure Retail Prices API, Google Cloud Billing Catalog, OCI Cost Estimator price list. 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, Oracle Cloud Infrastructure.
AWS
Azure
Google Cloud
Oracle