Best NeocloudsAn independent guide to GPU clouds

Home / GPU cloud prices, compared

GPU cloud prices, compared

148 published on-demand prices from 29 GPU clouds, per GPU-hour, checked on November 9, 2026. Every price links to the provider's own pricing page.

Last reviewed November 9, 2026. 21 sources linked at the end.

How to read these tables

Prices are in US dollars per GPU per hour, copied from each provider's pricing page on November 9, 2026. "Lowest" and "median" use fixed on-demand prices only. Marketplace offers and "from" prices are listed but left out of those two figures, because they depend on host, configuration or availability.

Reserved and committed prices are usually lower and almost never published. If you need more than a few nodes for more than a month, a quote will beat these numbers. Prices are rising for some GPUs; check the provider profile for announced changes.

NVIDIA H100

Lowest fixed on-demand price: $2.45 at Denvr Dataworks. Median across 20 fixed prices: $3.79.

ProviderGPUPer GPU-hourBasisSource
Latitude.shH100$1.68from, 1x GPU bare-metal server (g3.h100.small); form factor not statedPricing page
Voltage Park (Lightning AI)H100 SXM$1.99from; “starting at” figure in the pricing FAQ, no contract; plan cards say contact for pricingPricing page
Vast.aiH100 SXM$2.27marketplace, lowest live offer (median $2.87)Pricing page
Denvr DataworksH100 SXM$2.45on-demand, per GPU on 8x node, 3200G InfiniBand, billed per minutePricing page
Vast.aiH100 PCIe$2.47marketplace, lowest live offer (median $2.47)Pricing page
HyperstackH100 PCIe$2.50on-demand, listed as “NVIDIA H100” (NVLink variant $2.60)Pricing page
ShadeformH100 PCIe$2.50marketplace, lowest live offer (Hyperstack, 1x, Montreal)Pricing page
Vast.aiH100 NVL$2.53marketplace, lowest live offer (median $2.68)Pricing page
GMI CloudH100$2.60from; form factor and on-demand or committed not statedPricing page
Massed ComputeH100$2.73on-demand, 1x, 80GB, form factor not statedPricing page
RunpodH100 PCIe$2.89on-demand Pod, Secure Cloud, billed per secondPricing page
VultrH100 SXM$2.99on-demand list rate, per GPU; $23.92/hr 8-GPU HGX H100 plan (public plans API); 0 deployable locations at checkPricing page
Massed ComputeH100 NVL$3.11on-demand, 1xPricing page
Massed ComputeH100 SXM$3.14on-demand, 1x (8x = $25.12/hr)Pricing page
RunpodH100 NVL$3.19on-demand Pod, Secure Cloud, billed per secondPricing page
HyperbolicH100 SXM$3.19marketplace, from (best supplier rate, refreshed weekly)Pricing page
HyperstackH100 SXM$3.20on-demand, billed per minutePricing page
JarvislabsH100 SXM$3.49on-demand, billed per minute (USD rate outside India)Pricing page
Prime IntellectH100$3.50marketplace, on-demand price shown on compute page; form factor not statedPricing page
LambdaH100 PCIe$3.79on-demand, 1x instance; schedule effective 2026-11-16Pricing page
CrusoeH100 SXM$3.90on-demand, HGX H100; rises to $4.50 on 2026-12-01Pricing page
Together AIH100 SXM$3.99on-demand GPU cluster, pay as you go (HGX H100)Pricing page
RunpodH100 SXM$3.99on-demand Pod, Secure Cloud, billed per secondPricing page
DigitalOceanH100 SXM$4.41on-demand GPU Droplet (1x or 8x), billed per secondPricing page
NebiusH100 SXM$4.50on-demand, HGX H100Pricing page
LambdaH100 SXM$4.59on-demand, 8x HGX H100 instance; schedule effective 2026-11-16Pricing page
LambdaH100 SXM$4.89on-demand, 1x instance; schedule effective 2026-11-16Pricing page
VerdaH100 SXM$4.987on-demand, 1x instancePricing page
CoreWeaveH100 SXM$6.155on-demand, per GPU; $49.24/hr 8-GPU HGX H100 instancePricing page

NVIDIA H200

Lowest fixed on-demand price: $3.62 at Massed Compute. Median across 11 fixed prices: $5.29.

ProviderGPUPer GPU-hourBasisSource
GMI CloudH200$3.20from; on-demand or committed not statedPricing page
Vast.aiH200 NVL$3.33marketplace, lowest live offer (median $4.00)Pricing page
Vast.aiH200$3.43marketplace, lowest live offer (median $4.34)Pricing page
Massed ComputeH200 NVL$3.62on-demand, 1x (4x/8x NVLink configs $3.82 per GPU)Pricing page
HyperstackH200$3.99on-demand, H200 SXMPricing page
HyperbolicH200$3.99marketplace, from (best supplier rate, refreshed weekly)Pricing page
CrusoeH200$4.29on-demand; rises to $5.40 on 2026-12-01Pricing page
DigitalOceanH200$4.47on-demand GPU Droplet (1x or 8x), billed per secondPricing page
JarvislabsH200$4.59on-demand, H200 SXM, billed per minute (USD rate outside India)Pricing page
RunpodH200$5.29on-demand Pod, Secure Cloud, billed per secondPricing page
NebiusH200$5.40on-demandPricing page
Prime IntellectH200$5.48marketplace, on-demand price shown on compute page; offers vary by providerPricing page
Together AIH200$5.99on-demand GPU cluster, pay as you goPricing page
CoreWeaveH200$6.305on-demand, per GPU; $50.44/hr 8-GPU instancePricing page
RunpodH200$6.35on-demand Instant Cluster (H200 SXM); page lists $6.35/hr without restating the unitPricing page
VerdaH200$6.44on-demand, 1x instancePricing page

NVIDIA B200

Lowest fixed on-demand price: $5.43 at Massed Compute. Median across 10 fixed prices: $8.19.

ProviderGPUPer GPU-hourBasisSource
GMI CloudB200$5.00from; limited availability; on-demand or committed not statedPricing page
Massed ComputeB200$5.43on-demand, per GPU; $43.46/hr 8-GPU instancePricing page
Vast.aiB200$5.50marketplace, lowest live offer (median $8.75)Pricing page
HyperbolicB200$5.99marketplace, from (best supplier rate, refreshed weekly)Pricing page
HyperstackB200$6.00on-demand, billed per minutePricing page
LambdaB200$7.69on-demand, 8x HGX B200 instance; schedule effective 2026-11-16Pricing page
LambdaB200$7.99on-demand, 1x instance; schedule effective 2026-11-16Pricing page
RunpodB200$7.99on-demand Pod, Secure Cloud, billed per secondPricing page
Together AIB200$8.19on-demand GPU cluster, pay as you goPricing page
NebiusB200$8.50on-demandPricing page
VultrB200$8.50on-demand list rate, per GPU; $68.00/hr 8-GPU plan (public plans API); 0 deployable locations at checkPricing page
CoreWeaveB200$8.60on-demand, per GPU; $68.80/hr 8-GPU instancePricing page
VerdaB200$9.138on-demand, 1x instancePricing page

NVIDIA B300

Lowest fixed on-demand price: $6.60 at Massed Compute. Median across 7 fixed prices: $8.99.

ProviderGPUPer GPU-hourBasisSource
Massed ComputeB300$6.60on-demand, per GPU; $52.80/hr 8-GPU instancePricing page
Prime IntellectB300$6.64marketplace, on-demand price shown on compute page; offers vary by providerPricing page
HyperstackB300$7.40on-demand, billed per minutePricing page
Latitude.shB300$8.00from, per GPU; $64/hr 8-GPU bare-metal server (g4.b300.large)Pricing page
CrusoeB300$8.96on-demand; rises to $9.90 on 2026-12-01Pricing page
RunpodB300$8.99on-demand Pod, Secure Cloud, billed per secondPricing page
NebiusB300$9.50on-demandPricing page
Together AIB300$9.99on-demand GPU cluster, pay as you goPricing page
VerdaB300$11.81on-demand, 1x instancePricing page

NVIDIA GB200 NVL72

Lowest fixed on-demand price: $10.50 at CoreWeave.

ProviderGPUPer GPU-hourBasisSource
GMI CloudGB200 NVL72$8.00from; page does not say whether on-demand or committedPricing page
Vast.aiGB200$8.13marketplace, lowest live offer (median $9.06)Pricing page
CoreWeaveGB200 NVL72$10.50on-demand, per GPU; $42.00/hr 4-GPU instancePricing page

NVIDIA GB300 NVL72

Lowest fixed on-demand price: $13.58 at Verda.

ProviderGPUPer GPU-hourBasisSource
VerdaGB300$13.58on-demand, 1x GB300 instancePricing page

AMD Instinct MI300X and newer

Lowest fixed on-demand price: $2.59 at DigitalOcean. Median across 6 fixed prices: $3.80.

ProviderGPUPer GPU-hourBasisSource
DigitalOceanMI300X$2.59on-demand GPU Droplet (1x or 8x), billed per secondPricing page
VultrMI355X$2.59on-demand list rate, per GPU; $20.72/hr 8-GPU plan (public plans API); 0 deployable locations at checkPricing page
CrusoeMI300X$3.45on-demandPricing page
DigitalOceanMI325X$3.80on-demand GPU Droplet (1x or 8x), billed per secondPricing page
VultrMI300X$3.99on-demand list rate, per GPU; $31.92/hr 8-GPU bare metal (public plans API); 0 deployable locations at checkPricing page
VultrMI325X$4.615on-demand list rate, per GPU; $36.92/hr 8-GPU plan (public plans API); 0 deployable locations at checkPricing page

NVIDIA A100

Lowest fixed on-demand price: $0.89 at Jarvislabs. Median across 18 fixed prices: $1.79.

ProviderGPUPer GPU-hourBasisSource
Vast.aiA100 80GB$0.33marketplace, lowest live offer, PCIe (median $0.73)Pricing page
Vast.aiA100 80GB$0.37marketplace, lowest live offer, SXM4 (median $0.84)Pricing page
JarvislabsA100 40GB$0.89on-demand, billed per minutePricing page
Denvr DataworksA100 40GB$1.15on-demand, SXM, per GPU on 8x node, 800G InfiniBandPricing page
Denvr DataworksA100 40GB$1.15on-demand, PCIe, 4xPricing page
Massed ComputeA100 80GB$1.35on-demand, 1xPricing page
Massed ComputeA100$1.38on-demand, SXM, 1x (8x = $11.04/hr)Pricing page
ShadeformA100 80GB$1.38marketplace, lowest live offer, SXM4 (Massed Compute, 1x)Pricing page
Denvr DataworksA100 80GB$1.45on-demand, SXM, per GPU on 8x node, 1600G InfiniBandPricing page
JarvislabsA100 80GB$1.49on-demand, billed per minutePricing page
HyperstackA100$1.60on-demand, A100 SXMPricing page
RunpodA100$1.79on-demand Pod, Secure Cloud, SXMPricing page
RunpodA100$1.79on-demand Pod, Secure Cloud, PCIePricing page
CrusoeA100 80GB$2.00on-demand, PCIePricing page
RunpodA100$2.15on-demand Instant Cluster (A100 SXM); page lists $2.15/hr without restating the unitPricing page
CrusoeA100 80GB$2.30on-demand, SXMPricing page
VerdaA100 80GB$2.316on-demand, 1x instance, SXM4Pricing page
VultrA100 80GB$2.397on-demand list rate, 1x Cloud GPU (vGPU) instance (public plans API)Pricing page
CoreWeaveA100 80GB$2.70on-demand, per GPU; $21.60/hr 8-GPU instancePricing page
VultrA100$2.80on-demand list rate, per GPU; $22.40/hr 8-GPU bare metal, available in New Jersey (public plans API)Pricing page
LambdaA100 80GB$3.19on-demand, 8x HGX A100 instance; schedule effective 2026-11-16Pricing page

NVIDIA L40S

Lowest fixed on-demand price: $0.97 at Massed Compute. Median across 7 fixed prices: $1.57.

ProviderGPUPer GPU-hourBasisSource
Vast.aiL40S$0.47marketplace, lowest live offer (median $0.80)Pricing page
Massed ComputeL40S$0.97on-demandPricing page
RunpodL40S$1.09on-demand Pod, Secure Cloud, billed per secondPricing page
CrusoeL40S$1.50on-demandPricing page
ShadeformL40S$1.50marketplace, lowest live offer, per GPU; $15.00/hr 10-GPU VM (Crusoe, Culpeper VA)Pricing page
NebiusL40S$1.55from, on-demand (Intel CPU host; AMD CPU host from $1.82)Pricing page
DigitalOceanL40S$1.57on-demand GPU Droplet (1x), billed per secondPricing page
VultrL40S$1.671on-demand list rate, 1x Cloud GPU instance (public plans API); 0 deployable locations at checkPricing page
VerdaL40S$1.958on-demand, 1x instancePricing page
CoreWeaveL40S$2.25on-demand, per GPU; $18.00/hr 8-GPU instancePricing page

Which GPU should you rent?

For most fine-tuning and inference, H100 or H200 is still the price-performance default. B200 makes sense when you need the extra memory and FP4/FP8 throughput for large models. Our H100 vs H200 vs B200 guide compares specs and cost, and how to choose a GPU cloud covers what matters beyond the hourly price.

Common questions

What is the cheapest H100 per hour?

On November 9, 2026, the lowest fixed on-demand H100 price we found was $2.45 per GPU-hour at Denvr Dataworks (H100 SXM). Marketplace offers can be lower but change by the minute.

Are these the prices I will actually pay?

They are list prices for on-demand use. Reserved clusters, longer commitments and negotiated deals are usually cheaper, and those prices are rarely published. Taxes, storage and egress can add to the bill.

Why is H100 SXM more expensive than H100 PCIe?

SXM GPUs connect to each other over NVLink at much higher bandwidth and run at a higher power limit, which matters for multi-GPU training. PCIe cards are cheaper and fine for many single-GPU and inference jobs.

What does "from" mean next to a price?

The provider publishes a starting price that may depend on configuration, term or availability. We keep it in the table but leave it out of the lowest and median figures.

What is a marketplace price?

Marketplaces like Vast.ai list GPUs from many hosts. We show the lowest live offer on the review date. Hardware, network and reliability vary by host.

How often are these prices updated?

We recheck published prices at least monthly, and sooner when a provider announces a change. The date at the top of the page is the last full check.

Sources

  1. CoreWeave: pricing
  2. Nebius: pricing
  3. Lambda: pricing
  4. Crusoe: pricing
  5. Together AI: pricing
  6. Runpod: pricing
  7. Hyperstack: pricing
  8. Verda: pricing
  9. GMI Cloud: pricing
  10. Jarvislabs: pricing
  11. DigitalOcean: pricing
  12. Vultr: pricing
  13. Vultr: pricing
  14. Voltage Park (Lightning AI): pricing
  15. Denvr Dataworks: pricing
  16. Massed Compute: pricing
  17. Prime Intellect: pricing
  18. Vast.ai: pricing
  19. Shadeform: pricing
  20. Hyperbolic: pricing
  21. Latitude.sh: pricing