Compute
NVIDIA H200
As of 2026-11-15, renting one NVIDIA H200 on demand costs $3.59 to $10.60 per hour across 8 providers. The lowest price of any kind is $2.58 (CoreWeave, Spot). Source: AuraMatrix; list and market prices, not invoices.
Every price, cheapest first
| Provider | Kind | Per GPU-hour | Details | Course | Seen |
|---|---|---|---|---|---|
| CoreWeaveGPU cloud · HGX H200 | Spot | 2.58 | 8 GPUs in the machine at $20.64 per hourper GPU: 16 vCPUs and 256 GB memoryread from the page; two independent readers agree | 2026-11-15 | |
| Together AIGPU cloud | Spot | 2.99 | read from the page; two independent readers agree | · | 2026-11-12 |
| RunPod Community CloudCommunity hosts · SXM | On demand | 3.59 | from a machine-readable price list | · | 2026-11-15 |
| RunPod Secure CloudGPU cloud · NVL | On demand | 3.79 | from a machine-readable price list | · | 2026-11-15 |
| AWSCloud · p5en.48xlarge | Spot | 4.264 | 8 GPUs in the machine at $34.12 per hourper GPU: 24 vCPUs and 256 GB memorylowest in us-east-2b; 8 zones, up to $5.259from a machine-readable price list | 2026-11-15 | |
| Vast.aiMarketplace | Market (median ask) | 4.841 | from a machine-readable price list18 asks, from $1.713 to $8.291 | 2026-11-15 | |
| Microsoft AzureCloud · ND96isr_H200_v5 | Reserved, 3 years | 5.278 | 8 GPUs in the machine at $42.22 per hourper GPU: 12 vCPUs and 231.3 GB memorylowest in eastus2; 31 regions, up to $10.56from a machine-readable price list | 2026-11-15 | |
| RunPod Secure CloudGPU cloud · SXM | On demand | 5.29 | from a machine-readable price list | · | 2026-11-15 |
| Microsoft AzureCloud · ND96isr_H200_v5 | Reserved, 1 year | 5.815 | 8 GPUs in the machine at $46.52 per hourper GPU: 12 vCPUs and 231.3 GB memorylowest in eastus2; 31 regions, up to $11.63from a machine-readable price list | 2026-11-15 | |
| Together AIGPU cloud | On demand | 5.99 | read from the page; two independent readers agree | · | 2026-11-12 |
| CoreWeaveGPU cloud · HGX H200 | On demand | 6.305 | 8 GPUs in the machine at $50.44 per hourper GPU: 16 vCPUs and 256 GB memoryread from the page; two independent readers agree | 2026-11-15 | |
| AWSCloud · p5en.48xlarge | On demand | 7.912 | 8 GPUs in the machine at $63.30 per hourper GPU: 24 vCPUs and 256 GB memoryfrom a machine-readable price list | 2026-11-15 | |
| Google CloudCloud · a3-ultragpu-8g | On demand | 10.44 | 8 GPUs in the machine at $83.49 per hourper GPU: 28 vCPUs and 369 GB memoryGPUs, vCPUs and memory priced separately and added up; local SSD not includedlowest in us-central1; 6 regions, up to $12.52from a machine-readable price list | 2026-11-15 | |
| Microsoft AzureCloud · ND96isr_H200_v5 | On demand | 10.60 | 8 GPUs in the machine at $84.80 per hourper GPU: 12 vCPUs and 231.3 GB memorylowest in eastus2; 31 regions, up to $21.20from a machine-readable price list | 2026-11-15 | |
| Microsoft AzureCloud · ND96isr_H200_v5 | Spot | 10.60 | 8 GPUs in the machine at $84.80 per hourper GPU: 12 vCPUs and 231.3 GB memorylowest in eastus2; 31 regions, up to $21.20from a machine-readable price list | 2026-11-15 |
A cloud that sells whole machines is divided by its GPU count, so its price also covers the CPUs, memory and disk of the machine. A market price is the median of the asks on offer when the snapshot was taken.
Same GPU, different machine
Think of the GPU as the stove in a rented kitchen. The stove is the same everywhere. What differs is how many cooks prepare food for it (the vCPUs) and how much counter space there is (the memory). A strong stove with one cook and a small counter still serves slowly, so two prices for the same GPU are not always prices for the same thing.
| Provider | Per GPU-hour | vCPUs per GPU | Memory per GPU | Network | GPUs in the machine |
|---|---|---|---|---|---|
| CoreWeaveHGX H200 | 6.305 | 16 | 256 GB | — | 8 |
| AWSp5en.48xlarge | 7.912 | 24 | 256 GB | 3200 GbpsEFA | 8 |
| Google Clouda3-ultragpu-8g | 10.44 | 28 | 369 GB | 3600 Gbps | 8 |
| Microsoft AzureND96isr_H200_v5 | 10.60 | 12 | 231.3 GB | 80 Gbps | 8 |
Not stated on the price list or page of: RunPod Community Cloud, RunPod Secure Cloud, Together AI. Their rows above carry a price only. Network is the road to the kitchen: it decides how well several machines can work as one, which matters for training across many GPUs. These are the figures the providers state, not measurements.
The market, hour by hour
A list price is a menu. A marketplace is a street market: stalls come and go, and the going price moves through the day. We look once an hour and write down what is on offer.
| Variant | Median ask now | Median, hour by hour | Median ranged | Lowest ask now | Asks |
|---|---|---|---|---|---|
| H200 | 4.606 | 3.692 – 5.789 | 1.713 | 2052 GPUs |
Can you get one right now?
What RunPod’s API says it has in stock for one GPU, asked once an hour. A price on a list does not mean a machine is free. This is the provider’s own statement, not a test purchase; recording began 2026-11-13.
- H200 NVLlowof 68 hourly checks: low 5, none to rent 63
- H200 SXMmediumof 68 hourly checks: high 10, medium 22, low 36
Check a priceHolding a quote for this GPU?See where your price stands →
In your AIAsk your assistant: “What does an H200 cost on demand at each provider?”Connect in one click →
Sources
CoreWeave: pricing page (GPU instances)
- URL
- https://www.coreweave.com/pricing
- Fetched
- 2026-11-13 03:44 UTC
- SHA-256
- 8c0aae555514038f65be6b94cff772f992489bde9baf1dcb3bd933c48526168e
Together: pricing page (GPU clusters)
- URL
- https://www.together.ai/pricing
- Fetched
- 2026-11-12 05:03 UTC
- SHA-256
- 9d816b96aba8aa59462e12ab0fbd1d8cdd516c54e5417da02d1fe4f56d4e3daf
RunPod API: GPU types and prices
- URL
- https://api.runpod.io/graphql
- Fetched
- 2026-11-11 08:38 UTC
- SHA-256
- 87bad28672fc2bc93295946259c0d35144d2dbcf56db6ee2645858c6c493a4c2
AWS APIs: EC2 instance types, Price List (us-east-1) and spot prices
- URL
- https://docs.aws.amazon.com/AWSEC2/latest/APIReference/API_DescribeSpotPriceHistory.html
- Fetched
- 2026-11-15 22:16 UTC
- SHA-256
- 16 files; first 1d2d04d9cb57b738a1dc244102acc02a83b8a195e484f7fd744d480a48bd7062
Vast.ai marketplace: on-demand offers
- URL
- https://console.vast.ai/api/v0/bundles/
- Fetched
- 2026-11-15 22:15 UTC
- SHA-256
- 8 files; first a705ddbf75b31e7217a6c8b4be6744deb7a3efd7e8fb00cba46a5175e566452a
Azure Retail Prices API: GPU virtual machines
- URL
- https://prices.azure.com/api/retail/prices
- Fetched
- 2026-11-13 05:18 UTC
- SHA-256
- 15 files; first 958888fc5fd92479705ac6785ec6c657356c45863bbfdc29674958c2d0dd0fa1
AWS APIs: EC2 instance types, Price List (us-east-1) and spot prices
- URL
- https://docs.aws.amazon.com/AWSEC2/latest/APIReference/API_DescribeSpotPriceHistory.html
- Fetched
- 2026-11-13 05:18 UTC
- SHA-256
- 13 files; first 82619f5a6dd22c9d6b56082ac0613fb225e51d5232c7cb21e1afee5c28ae51e7
Google Cloud Billing Catalog API (Compute Engine) and the GPU machine types page
- URL
- https://cloud.google.com/compute/docs/gpus
- Fetched
- 2026-11-13 05:18 UTC
- SHA-256
- 2 files; first 9900a527b0481d6d3902059ce74fa6aedf168ed367162d67c9ee1dd68aa812b7