Compare providers
Sixteen providers, nineteen parts, one pricing type. On the NVIDIA H100 SXM the cheapest tracked provider sits 5.9× below the most expensive; on the NVIDIA H200 NVL the entire market fits inside 1.2×.
Each cell is that provider’s median on-demand price for the SKU today, per single GPU per hour. The lowest price in each row is highlighted — that spread is why this site exists. Spot and community tiers live on each SKU page.
Scroll sideways to compare providers →
| SKU |  Azure |  CoreWeave |  Verda |  Denvr |  Hot Aisle |  Hyperstack |  Lambda |  Massed Compute |  Nebius |  OVHcloud |  Vast.ai |  Voltage Park |  Vultr | Spread |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NVIDIA B300 | — | — | $9.15 | — | — | $6.00 | — | $6.60 | — | — | $15.00 | — | — | 2.5× |
| NVIDIA B200 | — | $8.60 | $7.27 | — | — | — | $6.89 | $5.43 | — | — | $7.81 | — | — | 1.6× |
| NVIDIA H200 SXM | — | $6.30 | $5.02 | — | — | — | — | — | $8.50 | — | — | — | — | 1.7× |
| NVIDIA H200 NVL | — | — | — | — | — | — | — | $3.72 | — | — | $4.51 | — | — | 1.2× |
| NVIDIA H100 SXM | $11.68 | $6.16 | $3.85 | $3.87 | — | $3.99 | $4.14 | $3.14 | $5.40 | — | $2.25 | $1.99 | — | 5.9× |
| NVIDIA H100 NVL | $6.98 | — | — | — | — | — | — | $3.01 | — | — | $3.02 | — | — | 2.3× |
| AMD Instinct MI300X | — | — | — | — | $2.99 | — | — | — | — | — | — | — | — | — |
| NVIDIA H100 PCIe | — | — | — | — | — | $2.60 | $3.29 | — | — | — | $2.87 | — | — | 1.3× |
| NVIDIA RTX PRO 6000 Blackwell | — | — | $2.13 | — | — | $2.50 | — | $2.19 | $4.50 | — | $1.44 | — | — | 3.1× |
| NVIDIA A100 80GB SXM | $4.10 | — | $1.85 | $2.45 | — | $1.85 | $2.79 | $1.38 | — | — | $1.17 | — | — | 3.5× |
| NVIDIA L40S | — | $2.25 | $1.59 | — | — | — | — | $0.97 | — | $1.80 | $0.71 | — | $1.67 | 3.2× |
| NVIDIA A100 40GB SXM | $3.40 | — | $1.35 | $1.45 | — | — | $1.99 | — | — | — | $0.91 | — | — | 3.7× |
| NVIDIA A100 80GB PCIe | $3.67 | — | — | — | — | $1.40 | — | — | — | — | $1.00 | — | — | 3.7× |
| NVIDIA L40 | — | $1.25 | — | — | — | $1.35 | — | — | — | — | $0.47 | — | — | 2.9× |
| NVIDIA A100 40GB PCIe | — | — | — | $1.15 | — | — | $1.99 | — | — | — | $0.63 | — | — | 3.1× |
| NVIDIA RTX 6000 Ada | — | — | — | — | — | — | — | — | — | — | $0.79 | — | — | — |
| NVIDIA RTX A6000 | — | — | — | — | — | $1.00 | $1.09 | $0.57 | — | — | $0.40 | — | — | 2.7× |
| NVIDIA RTX 5090 | — | — | — | — | — | — | — | — | — | — | $0.67 | — | — | — |
| NVIDIA RTX 4090 | — | — | — | — | — | — | — | — | — | — | $0.51 | — | — | — |
Multi-config providers (1x/2x/4x/8x instances) show their median across configs. Prices are as listed at scrape time; availability isn’t guaranteed. Links marked ↗ may be affiliate links — they never affect prices shown or ordering.
Tracked providers
- AzureVisitHyperscaler5 SKUs11,280 obs2026-10-08 00:11 UTC
- CoreWeaveVisitDedicated GPU cloud7 SKUs6,858 obs2026-10-08 00:11 UTC
- VerdaVisitDedicated GPU cloud · formerly DataCrunch8 SKUs36,146 obs2026-10-08 00:11 UTC
- DenvrVisitDedicated GPU cloud4 SKUs2,248 obs2026-10-08 00:11 UTC
- Hot AisleVisitDedicated GPU cloud1 SKUs562 obs2026-10-08 00:11 UTC
- HyperstackVisitDedicated GPU cloud8 SKUs4,520 obs2026-10-08 00:11 UTC
- LambdaVisitDedicated GPU cloud7 SKUs10,220 obs2026-10-08 00:12 UTC
- Massed ComputeVisitResells rented capacity9 SKUs6,182 obs2026-10-08 00:11 UTC
- NebiusVisitDedicated GPU cloud4 SKUs2,186 obs2026-10-08 00:11 UTC
- OVHcloudVisitDedicated GPU cloud1 SKUs1,689 obs2026-10-08 00:11 UTC
RunPodVisitMarketplace of independent hosts19 SKUs20,260 obs2026-10-05 18:24 UTC
SaladCloudVisitMarketplace of independent hosts2 SKUs836 obs2026-09-09 18:34 UTC
Thunder ComputeVisitResells rented capacity2 SKUs1,046 obs2026-09-29 01:55 UTC- Vast.aiVisitMarketplace of independent hosts19 SKUs58,176 obs2026-10-08 00:11 UTC
- Voltage ParkVisitDedicated GPU cloud1 SKUs545 obs2026-10-08 00:11 UTC
- VultrVisitDedicated GPU cloud1 SKUs1,671 obs2026-10-08 00:11 UTC
| Provider | What kind of supply | SKUs | Observations | Last seen | |
|---|---|---|---|---|---|
| Azure | Hyperscaler · US East list price; enterprise agreements and reservations are not tracked. | 5 | 11,280 | 2026-10-08 00:11 UTC | Visit |
| CoreWeave | Dedicated GPU cloud · Enterprise contracts are negotiated; the listed rate is the public one. | 7 | 6,858 | 2026-10-08 00:11 UTC | Visit |
| Verdaformerly DataCrunch | Dedicated GPU cloud | 8 | 36,146 | 2026-10-08 00:11 UTC | Visit |
| Denvr | Dedicated GPU cloud · On-demand table only; reserved pricing is excluded. | 4 | 2,248 | 2026-10-08 00:11 UTC | Visit |
| Hot Aisle | Dedicated GPU cloud · AMD Instinct specialist. | 1 | 562 | 2026-10-08 00:11 UTC | Visit |
| Hyperstack | Dedicated GPU cloud | 8 | 4,520 | 2026-10-08 00:11 UTC | Visit |
| Lambda | Dedicated GPU cloud | 7 | 10,220 | 2026-10-08 00:12 UTC | Visit |
| Massed Compute | Resells rented capacity | 9 | 6,182 | 2026-10-08 00:11 UTC | Visit |
| Nebius | Dedicated GPU cloud · Committed-use discounts are not tracked; on-demand only. | 4 | 2,186 | 2026-10-08 00:11 UTC | Visit |
| OVHcloud | Dedicated GPU cloud | 1 | 1,689 | 2026-10-08 00:11 UTC | Visit |
| RunPod | Marketplace of independent hosts · Secure Cloud and Community Cloud priced separately; we track them as different tiers. | 19 | 20,260 | 2026-10-05 18:24 UTC | Visit |
| SaladCloud | Marketplace of independent hosts · Consumer cards on idle household machines; interruption is the norm. | 2 | 836 | 2026-09-09 18:34 UTC | Visit |
| Thunder Compute | Resells rented capacity · Virtualized GPUs over the network. | 2 | 1,046 | 2026-09-29 01:55 UTC | Visit |
| Vast.ai | Marketplace of independent hosts · Peer-to-peer hosts bidding for work — the cheapest tier, with the weakest guarantees. | 19 | 58,176 | 2026-10-08 00:11 UTC | Visit |
| Voltage Park | Dedicated GPU cloud · Flat published rate, no negotiated tier. | 1 | 545 | 2026-10-08 00:11 UTC | Visit |
| Vultr | Dedicated GPU cloud · Whole-GPU plans only — fractional vGPU tiers are excluded. | 1 | 1,671 | 2026-10-08 00:11 UTC | Visit |
Missing a provider you use? Request it. Each one is an independent adapter — and sources that can’t give us a number we can stand behind stay out until they can (see methodology).