Try
10 GPU-hours
$23.20
$2.32/hour
- Test your setup end to end
- Same GPU and image as a long run
- Unused hours never expire
Quick answer
How much does it cost to rent an H100 per hour?
As of 29 October 2026, renting an on-demand NVIDIA H100 80GB costs $2.73 to $12.29 per GPU-hour across 16 providers. The median is $4.16. On GPUwithCrypto an H100 costs $2.32/hour, billed per minute and paid in crypto with no KYC. Spot and marketplace H100s go lower, from $0.79, but they can be interrupted.
How much does an H100 cost per hour at each provider?
One row per provider at its cheapest on-demand H100 configuration, sorted by price. According to GPUwithCrypto's price check of 29 October 2026, the most expensive on-demand H100 costs 4.5 times the cheapest. Azure charges $12.29 per GPU in an 8-GPU VM; Massed Compute charges $2.73.
| Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC |
|---|---|---|---|---|---|---|
| GPUwithCrypto | H100 80GB, 1 to 8 GPUs; SXM or PCIe shown per host | $2.32 | — (on-demand only) | Per minute | Yes | No |
| Massed Compute | H100 80GB (form factor not stated), 1 GPU | $2.73 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented |
| RunPod | H100 PCIe, 1 GPU (H100 SXM: $3.99) | $2.89 | $1.99 (Community Cloud) | Per second | Yes | KYC before first crypto payment |
| OVHcloud | H100 80GB (form factor not stated), 1 GPU | $2.99 | — | Per second, 1-minute minimum | No | Sometimes |
| Hyperbolic | H100 SXM | $3.19 | — | — | Unknown | None documented |
| Thunder Compute | H100 SXM, 1 GPU | $3.20 | — | Per minute | No | None documented |
| Lambda | H100 PCIe, 1 GPU (H100 SXM: $3.99 to $4.29) | $3.29 | — | Per minute | No (cards only) | None documented |
| Jarvislabs | H100 SXM, 1 GPU | $3.49 | $1.54 (spot) | Per minute | No | None documented |
| Crusoe | H100 SXM (HGX) | $3.90 | Contact sales | Per hour or per second | No | None documented |
| DigitalOcean (Paperspace) | H100 SXM (HGX), 1 or 8 GPUs | $4.41 | — | Per second, 5-minute minimum | Yes (stablecoins, limited) | None documented |
| Nebius | H100 SXM (HGX) | $4.50 | From $0.79 (preemptible) | — | No | None documented |
| Verda (formerly DataCrunch) | H100 SXM, 1 GPU | $4.75 | $2.37 (spot) | — | No | None documented |
| Spheron | H100 80GB (form factor not stated) | $5.86 | $2.79 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented |
| CoreWeave | H100 SXM, 8-GPU instance ÷ 8 | $6.16 | $2.46 (spot) | — | Unknown | Unknown |
| AWS | H100, 1 GPU (p5.4xlarge) or 8 GPUs (p5.48xlarge) | $6.88 | — | — | No | None documented |
| Google Cloud | H100 SXM, 8-GPU VM ÷ 8 (a3-highgpu-8g) | $11.06 | $6.59 (spot) | Per second, 1-minute minimum | No | None documented |
| Microsoft Azure | H100, 8-GPU VM ÷ 8 (ND H100 v5) | $12.29 | $2.27 (spot) | — | No | None documented |
| Vast.ai | H100 SXM, marketplace | — | $1.73 (Vast.ai's public price feed, cheapest rentable offer) | Per second | Yes (BitPay, Crypto.com) | None documented |
| Akash Network | H100 SXM, decentralized marketplace | — | $2.04 (lowest recent provider bid) | — | Yes (AKT, ACT) | None documented |
| Shadeform | H100 PCIe, 1 GPU, resold from other clouds | — | $2.50 (cheapest available listing) | — | Yes (USDC, per Stripe) | None documented |
| Voltage Park | H100 SXM (HGX) | Excluded: FAQ says 'starting at' $1.99, pricing cards say 'Contact for pricing' | — | — | Unknown | Unknown |
Prices checked on
USD per GPU-hour, taxes excluded, rounded to the cent. Hyperscaler prices cover the whole VM (AWS US East, Azure East US, Google Cloud Iowa). 'None documented' means no ID step in the official sign-up, billing or terms pages we read, not a promise that a provider never asks. RunPod's Community Cloud is community-hosted. Spheron's own H100 page said 'from $2.65/hr' the same day; its pricing-page figure is shown.
Why are spot and marketplace H100s cheaper than ours?
Because they can be interrupted, or because an independent host sets the price and runs the machine. Our price is set against a named on-demand reference, not against these.
- Preemptible: Nebius lists H100s from $0.79 per GPU-hour. Nebius can reclaim them, and the rate floats, capped one cent below its on-demand price.
- Spot: Jarvislabs sells interruptible H100 SXM hours at $1.54.
- Marketplace: Vast.ai's public price feed showed 56 rentable H100 SXM offers, from $1.73 with a median of $2.40. The feed does not say which offers are interruptible.
- Community-hosted: RunPod's Community Cloud listed an H100 PCIe at $1.99, against $2.89 on its Secure Cloud.
A job that checkpoints every few minutes can save money on spot. A long training run or a live inference server needs a GPU that stays put. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it.
Which H100 sets the reference price, and why is Voltage Park's $1.99 left out?
The reference is Massed Compute's 'H100 (80GB)' at $2.73 per GPU-hour, and its page does not state the form factor. It could be SXM or PCIe. If you need an SXM card, compare with SXM prices: across 12 self-serve providers the lowest was $3.14 (Massed Compute) and the median $3.99.
Voltage Park is left out because its on-demand pricing cards say 'Contact for pricing'. The $1.99 figure is a 'starting at' line in its FAQ, not a price you can pay without talking to sales.
GPUwithCrypto has one H100 80GB price for every host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect.
How much does an H100 cost per day, week or month?
Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same hours.
| Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere |
|---|---|---|---|---|
| 1 hour | 1 | $2.32 | $2.73 | $4.16 |
| 1 day | 24 | $55.68 | $65.52 | $99.72 |
| 1 week | 168 | $389.76 | $458.64 | $698.04 |
| 30 days | 720 | $1,670.40 | $1,965.60 | $2,991.60 |
Prices checked on
The median is $4.155 per GPU-hour before rounding. Our day, week and month match the 24, 168 and 720 GPU-hour blocks.
What are the H100's specs?
NVIDIA's official figures for the three H100 versions sold in the cloud. Compare dense with dense: NVIDIA prints H100 Tensor figures with sparsity only, which doubles peak throughput for 2:4 sparse weights.
| Spec | H100 SXM | H100 PCIe | H100 NVL |
|---|---|---|---|
| Memory | 80 GB HBM3 | 80 GB HBM2e | 94 GB HBM3 |
| Memory bandwidth | 3,350 GB/s | 2,000 GB/s | 3,900 GB/s |
| FP8 Tensor, dense | 1,979 TFLOPS (3,958 sparse) | Not on NVIDIA's current page | 1,670.5 TFLOPS (3,341 sparse) |
| FP16 / BF16 Tensor, dense | 989.5 TFLOPS (1,979 sparse) | Not on NVIDIA's current page | 835.5 TFLOPS (1,671 sparse) |
| GPU-to-GPU link | NVLink 900 GB/s | NVLink bridge 600 GB/s | NVLink bridge 600 GB/s |
| Max power | 700 W | 350 W | 400 W |
Dense = NVIDIA's sparse figure ÷ 2. Almost no off-the-shelf LLM checkpoint is 2:4 sparse, so the dense figure matches real training and inference.
What fits on one H100?
Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb) and weights plus optimizer and gradients for fine-tuning. A full 128K-token context for Llama 3.1 70B adds another 42.9 GB of BF16 KV cache. The VRAM guide has the formulas.
| Model and job | Memory needed | One H100 80GB? |
|---|---|---|
| Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | ✓ |
| Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | ✓ |
| Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | No: 2× 80 GB or more |
| Llama 3.1 70B inference, 4-bit | ≈ 42 GB | ✓ |
| Llama 3.1 70B QLoRA fine-tune | ≈ 52.5 GB | ✓ |
| Llama 3.1 70B inference, FP8 | ≈ 84 GB | No: just over 80 GB; one H100 NVL (94 GB) or one H200 |
| Llama 3.1 70B inference, BF16 | ≈ 168 GB | No: 3× H100 |
| Qwen2.5 72B inference, 4-bit | ≈ 43.6 GB | ✓ |
| FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | ✓ |
| DeepSeek-V3 or R1 inference, FP8 | ≈ 805 GB | No: 11× 80 GB |
Which H100 providers accept crypto?
Of the 16 self-serve H100 providers above, 4 accept crypto: Massed Compute (BTC), RunPod, DigitalOcean (stablecoins, limited availability) and Spheron (USDC). RunPod's billing docs say to complete any required KYC before the first crypto payment. The marketplaces Vast.ai and Akash also take crypto.
GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, from a $25 top-up, with no ID, no selfie and no card. The H100 with crypto page compares the crypto options step by step.
What does an H100 cost on GPUwithCrypto?
$2.32/hour, at least 15% under its named on-demand reference: Massed Compute at $2.73, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire.
Popular
Day
24 GPU-hours
$55.68
$2.32/hour
- One day of continuous running
- A fine-tuning run or a batch job
- Unused hours never expire
Week
168 GPU-hours
$389.76
$2.32/hour
- One week of continuous running
- Training runs and evaluations
- Unused hours never expire
Month
720 GPU-hours
$1,670.40
$2.32/hour
- 30 days of continuous running
- An inference server running all month
- Unused hours never expire
In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire
H100 price questions
What is the cheapest H100 cloud right now?
On 29 October 2026 the lowest self-serve on-demand H100 price we found was $2.73 at Massed Compute, and GPUwithCrypto charges $2.32. Interruptible H100s cost less: Nebius preemptible from $0.79, and $1.73 in Vast.ai's public price feed.
Is an H100 SXM more expensive than an H100 PCIe?
Usually. The median H100 SXM cost $3.99 per GPU-hour across 12 self-serve providers, while the two PCIe prices were $2.89 (RunPod) and $3.29 (Lambda). The SXM has 3,350 GB/s of memory bandwidth against 2,000 GB/s.
How much does an H100 cost per month?
720 GPU-hours cost $1,670.40 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $2,991.60, and $1,965.60 at the lowest.
Can I rent an H100 with crypto and no KYC?
Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC with no ID, no selfie and no card. Elsewhere, 4 of the 16 self-serve H100 providers we checked accept crypto, and RunPod asks for KYC before the first crypto payment.
Should I rent an H100 or an A100?
Rent the A100 when your model fits in 80 GB and runs in BF16: it costs $0.92/hour here. Rent the H100 for FP8, which the A100 lacks, and for up to 64% more memory bandwidth on the SXM version. The H100 vs A100 comparison goes further.
Sources (24)
- Massed Compute — VM pricing vm.massedcompute.com/pricing — Massed Compute (read 2026-10-29)
- Runpod — Pricing runpod.io/pricing — RunPod (read 2026-10-29)
- OVHcloud — Public Cloud prices ovhcloud.com/en/public-cloud/prices — OVHcloud (read 2026-10-29)
- Hyperbolic — GPU marketplace hyperbolic.ai/marketplace — Hyperbolic (read 2026-10-29)
- Thunder Compute — Pricing thundercompute.com/pricing — Thunder Compute (read 2026-10-29)
- Lambda — Pricing lambda.ai/pricing — Lambda (read 2026-10-29)
- Jarvislabs — Pricing jarvislabs.ai/pricing — Jarvislabs (read 2026-10-29)
- Crusoe — Cloud pricing crusoe.ai/cloud/pricing — Crusoe (read 2026-10-29)
- DigitalOcean — GPU Droplets pricing digitalocean.com/pricing/gpu-droplets — DigitalOcean (read 2026-10-29)
- Nebius — Prices nebius.com/prices — Nebius (read 2026-10-29)
- Verda — Pricing verda.com/pricing — Verda (read 2026-10-29)
- Spheron — Pricing spheron.network/pricing — Spheron (read 2026-10-29)
- CoreWeave — Pricing coreweave.com/pricing — CoreWeave (read 2026-10-29)
- AWS — EC2 on-demand price list, US East (N. Virginia) — Amazon Web Services (read 2026-10-29)
- Google Cloud — Accelerator-optimized VM pricing — Google Cloud (read 2026-10-29)
- Azure Retail Prices API, East US virtual machines — Microsoft (read 2026-10-29)
- Vast.ai public GPU pricing feed — Vast.ai (read 2026-10-29)
- Akash Network — GPU prices API console-api.akash.network/v1/gpu-prices — Akash Network (read 2026-10-29)
- Shadeform — Instance types API api.shadeform.ai/v1/instances/types — Shadeform (read 2026-10-29)
- Voltage Park — Pricing voltagepark.com/pricing — Voltage Park (read 2026-10-29)
- Runpod — Billing information docs.runpod.io/references/billing-information — RunPod (read 2026-10-29)
- NVIDIA H100 Tensor Core GPU — NVIDIA (read 2026-10-29)
- NVIDIA H100 PCIe product brief — NVIDIA (read 2026-10-29)
- Transformer Math 101 — EleutherAI (read 2026-10-29)