Daily price tracker · November 2026
GPU Cloud Pricing: H100, H200, B200 and A100 Cost per Hour
Compare on-demand rental prices for NVIDIA data center GPUs across 31 cloud providers, from hyperscalers like AWS and Google Cloud to specialist GPU clouds. All prices are normalised to US dollars per GPU per hour and refreshed every day.
Prices by GPU and provider
On-demand prices, sorted from cheapest. Multi-GPU servers are divided by GPU count. Click the price column to change the sort order.
NVIDIA H100 80GB cloud prices per GPU-hour
| Provider | Price / GPU-hr ↕ | Configuration | Type | 8 GPUs / hr | 1 GPU / month |
|---|---|---|---|---|---|
| HyperAI Lowest | 1x H100 | GPU cloud | $14.40 | $1,314 | |
| Hyperstack | 1x H100 PCIe | GPU cloud | $20.00 | $1,825 | |
| UpCloud | 1x H100 SXM | GPU cloud | $20.00 | $1,825 | |
| Civo | 1x H100 SXM | GPU cloud | $23.92 | $2,183 | |
| OVHcloud | 1x H100 | GPU cloud | $23.92 | $2,183 | |
| Hyperbolic | 1x H100 | GPU cloud | $25.52 | $2,329 | |
| Scaleway | 1x H100 PCIe | GPU cloud | $25.52 | $2,329 | |
| Thunder Compute | 1x H100 SXM | GPU cloud | $25.60 | $2,336 | |
| Lambda | 1x H100 PCIe | GPU cloud | $26.32 | $2,402 | |
| Jarvislabs | 1x H100 SXM | GPU cloud | $27.92 | $2,548 | |
| Crusoe | 8x H100 SXM | GPU cloud | $31.20 | $2,847 | |
| Modal | 1x H100, serverless | Serverless / marketplace | $31.60 | $2,884 | |
| Together AI | 1x H100 SXM | GPU cloud | $31.92 | $2,913 | |
| RunPod | 1x H100 SXM, Secure Cloud | GPU cloud | $31.92 | $2,913 | |
| DigitalOcean | 1x H100 SXM | GPU cloud | $35.28 | $3,219 | |
| Nebius | 1x H100 SXM | GPU cloud | $36.00 | $3,285 | |
| Paperspace | 1x H100 SXM | GPU cloud | $47.60 | $4,344 | |
| CoreWeave | 8x H100 SXM | GPU cloud | $49.28 | $4,497 | |
| AWS | p5.4xlarge, 1x H100 | Hyperscaler | $55.04 | $5,022 | |
| Azure | NCads H100 v5, 1x H100 NVL | Hyperscaler | $55.84 | $5,095 | |
| Oracle Cloud | BM.GPU.H100.8, 8x H100 | Hyperscaler | $80.00 | $7,300 | |
| Google Cloud | a3-highgpu-8g, 8x H100 | Hyperscaler | $88.56 | $8,081 |
NVIDIA H200 141GB cloud prices per GPU-hour
| Provider | Price / GPU-hr ↕ | Configuration | Type | 8 GPUs / hr | 1 GPU / month |
|---|---|---|---|---|---|
| Beam Lowest | 1x H200 | Serverless / marketplace | $16.72 | $1,526 | |
| Ori | 1x H200 SXM | GPU cloud | $28.00 | $2,555 | |
| Vast.ai | Marketplace median, US/CA hosts | Serverless / marketplace | $30.80 | $2,811 | |
| Hyperbolic | 1x H200 | GPU cloud | $31.92 | $2,913 | |
| Crusoe | H200 | GPU cloud | $34.32 | $3,132 | |
| DigitalOcean | H200 | GPU cloud | $35.76 | $3,263 | |
| Modal | H200, serverless | Serverless / marketplace | $36.32 | $3,314 | |
| RunPod | H200 | GPU cloud | $42.32 | $3,862 | |
| Nebius | H200 | GPU cloud | $43.20 | $3,942 | |
| CoreWeave | 8x H200 | GPU cloud | $50.48 | $4,606 | |
| AWS | p5en.48xlarge, 8x H200 | Hyperscaler | $63.28 | $5,774 | |
| Oracle Cloud | BM.GPU.H200.8, 8x H200 | Hyperscaler | $80.00 | $7,300 | |
| Azure | ND96isr H200 v5, 8x H200 | Hyperscaler | $84.80 | $7,738 | |
| Google Cloud | a3-ultragpu-8g, reservation only | Hyperscaler | $86.80 | $7,921 |
NVIDIA B200 192GB cloud prices per GPU-hour
| Provider | Price / GPU-hr ↕ | Configuration | Type | 8 GPUs / hr | 1 GPU / month |
|---|---|---|---|---|---|
| Beam Lowest | 1x B200 SXM | Serverless / marketplace | $32.88 | $3,000 | |
| Hostinger | 1x B200 | GPU cloud | $36.00 | $3,285 | |
| GMI Cloud | 1x B200 | GPU cloud | $40.00 | $3,650 | |
| Massed Compute | 8x B200 SXM | GPU cloud | $43.44 | $3,964 | |
| Hyperstack | 8x B200 SXM | GPU cloud | $48.00 | $4,380 | |
| UpCloud | 1x B200 SXM | GPU cloud | $48.40 | $4,417 | |
| Lambda | 8x B200 SXM | GPU cloud | $53.52 | $4,884 | |
| RunPod | 1x B200, Secure Cloud | GPU cloud | $63.92 | $5,833 | |
| Together AI | 1x B200 SXM | GPU cloud | $65.52 | $5,979 | |
| Nebius | 1x B200 SXM | GPU cloud | $68.00 | $6,205 | |
| CoreWeave | 8x B200 SXM | GPU cloud | $68.80 | $6,278 | |
| Lightning AI | 8x B200 | Serverless / marketplace | $78.88 | $7,198 | |
| Oracle Cloud | 8x B200 | Hyperscaler | $112.00 | $10,220 | |
| AWS | p6-b200.48xlarge, 8x B200 | Hyperscaler | $113.92 | $10,395 |
NVIDIA A100 80GB / 40GB cloud prices per GPU-hour
| Provider | Price / GPU-hr ↕ | Configuration | Type | 8 GPUs / hr | 1 GPU / month |
|---|---|---|---|---|---|
| Jarvislabs Lowest | 1x A100 40GB | GPU cloud | $7.12 | $650 | |
| Thunder Compute | 1x A100 80GB | GPU cloud | $8.72 | $796 | |
| Seeweb | 1x A100 80GB PCIe | GPU cloud | $8.80 | $803 | |
| Hyperstack | 1x A100 80GB PCIe | GPU cloud | $10.80 | $986 | |
| Massed Compute | 1x A100 80GB SXM | GPU cloud | $11.04 | $1,007 | |
| RunPod | 1x A100 80GB SXM, Secure Cloud | GPU cloud | $14.32 | $1,307 | |
| Lambda | 1x A100 40GB SXM | GPU cloud | $15.92 | $1,453 | |
| Crusoe | 1x A100 80GB PCIe | GPU cloud | $16.00 | $1,460 | |
| Hugging Face | 1x A100 80GB, a100-large | Serverless / marketplace | $20.00 | $1,825 | |
| CoreWeave | 8x A100 80GB SXM | GPU cloud | $21.60 | $1,971 | |
| AWS | p4d.24xlarge, 8x A100 40GB | Hyperscaler | $21.92 | $2,000 | |
| Oracle Cloud | BM.GPU4.8, 8x A100 40GB | Hyperscaler | $24.40 | $2,227 | |
| OVHcloud | 1x A100 80GB | GPU cloud | $24.56 | $2,241 | |
| Azure | ND96asr A100 v4, 8x A100 40GB | Hyperscaler | $27.20 | $2,482 | |
| Google Cloud | a2-megagpu-16g, 16x A100 40GB | Hyperscaler | $29.76 | $2,716 |
"1 GPU / month" assumes 730 hours of continuous use. Prices exclude storage, networking and taxes. Availability varies, and some providers only rent full 8-GPU servers.
Daily price history
Median on-demand price per GPU-hour across tracked providers, last 14 days.
| Date | H100 | H200 | B200 | A100 |
|---|---|---|---|---|
| 2026-11-13 | $3.92 | $4.92 | $7.34 | $2.00 |
GPU rental cost calculator
Other GPUs: market snapshot
| GPU | Typical $/GPU-hr | Lowest tracked |
|---|---|---|
| NVIDIA B300 | $9.50 | $5.53 |
| NVIDIA GB200 | $13.25 | $8.00 |
| NVIDIA RTX PRO 6000 (96GB) | $2.38 | $0.65 |
| NVIDIA L40S (48GB) | $1.09 | $0.97 |
| NVIDIA RTX 5090 (32GB) | $0.72 | $0.35 |
| NVIDIA RTX 4090 (24GB) | $0.47 | $0.33 |
| AMD Instinct MI300X (192GB) | $2.59 | $2.59 |
| AMD Instinct MI325X | n/a | $2.45 |
What the numbers show today
Hyperscalers cost about 1.8x more
The cheapest on-demand H100 at a hyperscaler is $6.88 per GPU-hour, against a market median of $3.92. Big clouds close the gap with reservations and savings plans.
The H200 premium is about 16%
At providers that offer both, the H200 costs on average about 16% more than the H100, while giving 76% more memory (141 GB vs 80 GB) and faster bandwidth. For large models it is often better value.
B200 costs about 1.9x an H100
The median B200 rents for $7.34 per hour, but offers much higher FP8 and FP4 throughput and 192 GB of memory per GPU.
A100 is the budget option
An A100 rents from $0.89 per hour. It lacks FP8, but is still strong for fine-tuning and inference of smaller models.
How to choose a GPU cloud
- Match the GPU to the model. Check memory first with our guide to how much VRAM an LLM needs, then compare speed. See H100 vs H200 and B200 vs H100.
- Look at the full bill. Storage, data transfer (egress) and idle time can add a lot to the GPU price.
- Check availability. The cheapest listings are often out of stock or limited to single GPUs.
- Multi-GPU training needs fast links. For training across many GPUs, choose SXM servers with NVLink and InfiniBand networking. Learn why in NVLink and NVSwitch explained.
- Commit only when usage is steady. Reserved pricing saves money, but only if you will keep the GPUs busy.
Methodology and sources
Every day we compile on-demand list prices published by providers, as aggregated by public GPU price trackers. Prices are converted to US dollars per GPU per hour; multi-GPU instance prices are divided by the number of GPUs. The median is calculated across the providers listed for each GPU. We exclude spot, reserved and custom-quote pricing. Prices change frequently and may differ by region, so always confirm on the provider's site. Sources: GetDeploying GPU price data, GPU Per Hour price index and Thunder Compute H200 price comparison.
Frequently asked questions
How much does an H100 cost per hour?
As of November 13, 2026, the median on-demand price for an NVIDIA H100 across the 22 providers we track is $3.92 per GPU-hour. The lowest listing is $1.80 (HyperAI), and hyperscalers such as AWS, Azure, Oracle and Google Cloud charge about $6.88 to $11.07 per GPU-hour on demand.
How much does an H200 cost per hour?
The median on-demand rate is $4.92 per GPU-hour, with listings from $2.09 to $10.85. At providers that offer both, the H200 costs on average about 16% more per hour than the H100.
How much does a B200 cost per hour?
The median on-demand price is $7.34 per GPU-hour, ranging from $4.11 (Beam) to $14.24.
Why are AWS, Azure and Google Cloud more expensive?
Hyperscalers bundle GPUs with enterprise networking, security, compliance, support and integration with their other services, and they mostly rent full 8-GPU servers. Specialist GPU clouds focus on GPUs alone and compete hard on price. Hyperscaler prices drop a lot with reservations, savings plans or spot capacity.
What is the difference between on-demand, reserved and spot pricing?
On-demand means paying by the hour with no commitment, the prices shown here. Reserved or committed pricing gives large discounts for 1 to 3 year contracts. Spot or interruptible capacity is cheapest but can be taken back at short notice, so it suits jobs that can restart.
Is it cheaper to rent or buy a GPU?
Renting is cheaper unless you can keep GPUs busy most of the time for years. A data center GPU costs tens of thousands of dollars to buy, plus servers, power, cooling and networking. At the current median H100 price, a GPU rented around the clock costs about $2,862 per month, so buying only pays off with high, steady utilisation.
How often is this page updated?
Prices are refreshed every day and recorded in the price history table, so you can see how the market moves. Prices can change during the day, so always confirm the final price on the provider site before you rent.