Cloud GPU Pricing — Cost Intelligence for H100, A100 & B200
GPU pricing stopped being a stable lookup table. We track 16 providers, every major GPU, and every pricing tier independently, so your team has a neutral view before committing budget.
$2.50
Cheapest H100 /GPU/hr
Hyperstack · 64% below AWS
$1.56
Cheapest H100 spot /GPU/hr
Hyperstack H100 NVLink 80GB
16
Providers Tracked
Marketplace rates shown separately
49
GPU Models Tracked
Distinct models in this view
Tools
GPU price arbitrage by region
Pick a GPU model and see every tracked provider's on-demand rate, grouped by region and sorted cheapest to most expensive.
Cheapest
Azure in US East
$0.90/GPU-hr
Most expensive tracked
Lambda in US West
$1.29/GPU-hr
1.4x price difference ($0.39/GPU-hr gap)
US East
| Provider | Instance | $ / GPU-hr |
|---|---|---|
| Azure | NV18ads_A10_v5 | $0.90 |
NV18ads_A10_v5
US West
| Provider | Instance | $ / GPU-hr |
|---|---|---|
| Lambda | gpu_1x_a10 | $1.29 |
gpu_1x_a10
Prices as tracked; confirm current rates on the provider's page before procurement.
| Trend | Actions | Deploy | ||||
|---|---|---|---|---|---|---|
Atlantic.Net L40S 48GB · 48 GB | $1.67 −53% vs AWS | —Provider-reported, Atlantic.Net public p… | $1.65 | — | ||
Atlantic.Net H100 NVL 94GB · 94 GB | $3.94 −43% vs AWS | —Provider-reported, Atlantic.Net public p… | $3.79 | — |
Use Spot for Training
Save 60–70% on training runs with checkpointing. GCP Spot offers up to 70% discount on A100/H100 instances.
Best for: Fault-tolerant training with checkpoints
Reserved for Inference
1-year commitments save 35–40% for always-on inference endpoints. Azure and AWS offer the deepest reserved discounts.
Best for: Production inference workloads
Lambda/Nebius for Value
GPU-specialist clouds offer 2–4× lower per-GPU pricing than hyperscalers. Nebius leads on committed pricing; Lambda leads on on-demand.
Best for: Pure GPU compute without cloud services
Provider Directory
Hyperscaler. Capacity Blocks and Savings Plans for committed GPU capacity.
Hyperscaler. Offers NVIDIA GPUs and Google TPUs.
Hyperscaler. Offers NVIDIA and AMD Instinct GPU instances.
GPU-focused cloud with on-demand and reserved clusters.
GPU-focused cloud. Committed pricing is quoted per contract.
Secure Cloud (dedicated) and Community Cloud (shared) are separate tiers here.
GPU cloud with EU and US data centers.
Inference API and dedicated GPU clusters on one platform.
GPU cloud. Provider states no egress fees.
GPU cloud with European regions. Provider states no egress fees.
GPU cloud with per-minute billing.
GPU sandboxes for AI-generated code execution, billed per second.
Small-form-factor and workstation GPU hosting (UK, Portugal).
Marketplace. Rates are set by individual hosts and shown separately from dedicated-tier rates.
GPU node rental. Rates shown per node, normalized per GPU.
GPU cloud hosting with on-demand, 1-year, and 3-year terms.
Prices vary by region and availability. Verify with the provider before procurement. Rows show when they were last tracked.