Cheapest H100$2.59/hron Vast.ai
Cheapest H200$3.43/hron QuantaCloud
🌍Cheapest B300$7.89/hron RunPod
Tracked offers:19,732
Available:804

Compare cloud GPU prices

Live offers from 31 providers, priced per GPU-hour.

The market, at a glance

GPU market snapshot

Current secure on-demand prices across 15 popular GPUs. Every price is per GPU-hour.

Explore the price index
H100 SXM5$2.90/ GPU-hourOri · 5 providers with stockH200 NVL$3.43/ GPU-hourQuantaCloud · 3 providers with stockA100 SXM4 80GB$0.80/ GPU-hourVast.ai · 5 providers with stockL40S$0.80/ GPU-hourVast.ai · 6 providers with stock

Cheapest current secure on-demand offers · per GPU-hour

Latest stock observation:

Cheapest current GPU prices and historical catalog averages
GPUCurrent priceProviderHistorical catalog avg.Providers with stock
H100 SXM5all H100$2.90/hrOri$2.18/hr (30 days)5
H100 PCIeall H100$2.59/hrVast.ai$2.77/hr (30 days)8
H100 NVLall H100$3.11/hrMassed Compute$1.91/hr (30 days)3
H200 SXMall H200$3.50/hrOri$2.70/hr (30 days)4
H200 NVLall H200$3.43/hrQuantaCloud$2.89/hr (30 days)3
B200 SXMall B200$6.79/hrRunPod$4.98/hr (30 days)3
B300 SXM6all B300$7.89/hrRunPod$5.99/hr (30 days)3
A100 SXM4 80GBall A100$0.80/hrVast.ai$1.06/hr (30 days)5
A100 PCIe 80GBall A100$0.68/hrLeaderGPU$1.69/hr (30 days)7
L40S$0.80/hrVast.ai$0.90/hr (30 days)6
RTX PRO 6000 Blackwell MIG 24GB$0.59/hrRunPod$1.79/hr (30 days)6
RTX 5090$0.53/hrVast.ai$0.46/hr (30 days)3
RTX 4090$0.45/hrVast.ai$0.33/hr (30 days)3
RTX A6000$0.44/hrLeaderGPU$0.57/hr (30 days)6
MI300X$2.99/hrHot Aisle$2.78/hr (30 days)1

Historical averages include all catalog listings, including unavailable and spot offers. They do not measure historical live prices or savings. Availability can change before booking.

Daily history and CSV downloads: Cloud GPU Price Index

Sources and methodology

Current offers must be marked available by a provider with live stock tracking and observed within 15 minutes of the summary’s last refresh. The summary refreshes at most every 5 minutes and includes only secure on-demand instances, excluding spot and peer-to-peer hosts. Historical figures are medians of daily catalog averages over the last 30 days, including all pricing types and availability states. Missing days are excluded.

GPUPerHour scrapes the public pricing and inventory endpoints of 31 cloud GPU providers periodically. Each listing is normalised to a price per GPU per hour (multi-GPU instance price divided by GPU count) and tagged with its pricing type and security tier. The current-price column uses only secure on-demand listings; spot, reserved and peer-to-peer marketplace listings are tracked but excluded here because their prices are not comparable. The full list of tracked GPU models is on the GPU index.

A better starting point

Choose well. Spend less. Get to work.

Make sense of the specs, compare the complete bill, and rent the GPU that fits your workload.

Explore buying guides
01

How to compare cloud GPU pricing

Compare like-for-like GPUs, then account for storage, transfers, and pricing commitments.

Read the guide

Compare the same GPU model, memory size, pricing type and region before choosing a provider. Our current-price summary uses secure on-demand offers with recent stock observations. Each price is per GPU per hour: an eight-GPU instance priced at $24 per hour is $3 per GPU-hour, but the provider still bills $24 for the whole instance. Offer counts describe configurations, not the number of physical GPUs available to book.

On-demand, spot and reserved prices carry different commitments. Spot capacity may be interrupted; reserved offers may require a minimum term or payment. Storage, network transfer and other charges can change the total bill. Use the pricing tool's filters and cost calculator, and check the provider's terms. Our data egress reference explains network transfer charges.

02

Choose a GPU for your workload

Start with the memory your model needs, then check bandwidth, precision, and connectivity.

Read the guide

Start with the GPU memory your model, batch size and precision require. Then compare throughput, memory bandwidth and interconnect requirements using your own workload. Explore current offers for H100, H200, A100 80GB, A100 40GB, B200, and RTX 4090. The GPU catalog, comparison pages and provider profiles cover the wider market.

03

How to rent a cloud GPU by the hour

Choose an available offer, connect to your instance, and stop it when the work is done.

Read the guide

Renting a cloud GPU means provisioning a virtual machine, container or bare-metal server with one or more GPUs attached, paying for the time it runs, and releasing it when the job is done. There is no hardware to buy or maintain.

  1. Choose a GPU model whose memory fits your model, batch size and precision; the GPU catalog lists every tracked model with its current offers.
  2. Compare the per-GPU hourly price, pricing type, security tier and region across providers in the market summary and the pricing table above.
  3. Create an account with the chosen provider, or use DeployGPU to deploy on supported providers from one console.
  4. Provision the instance through the provider’s console, API or CLI and connect over SSH or a web terminal.
  5. Terminate the instance when the workload finishes to stop billing, and check the provider’s storage and network transfer charges.
From the research desk

Understand your next GPU decision.

Hardware explainers, practical workload guides, and pricing analysis with sources you can check.

All guides and analysis
hardware / guides

NVIDIA Competitors: Choose by Rental, Cloud or API

Compare AI chips by how you access them: AMD rentals, Google and AWS cloud accelerators, or Cerebras and Groq APIs. Get live prices and a clear decision rule.

12 min read
guides / pricing

AWS GPU Instances Decoded Across Four Major Clouds

Decode AWS, Google Cloud, Azure and Oracle GPU names, check GPU counts and memory, then compare live AWS rates and dated Oracle rates with other providers.

13 min read
guides / pricing

ComfyUI Cloud: Rent for Your Largest Model's VRAM

Match SDXL, FLUX and video workflows to GPU memory, compare live rental options, and weigh Comfy Cloud subscriptions against your monthly GPU usage.

13 min read
hardware / guides

HBM4 Memory Explained: HBM4 vs HBM3E and Which GPUs Use It

Compare HBM4 with HBM3E, separate JEDEC limits from maker claims, and check Rubin and AMD memory plans against the HBM3E GPUs in live rental tables.

13 min read
guides / analysis

LoRA vs QLoRA: Choose by VRAM, Then Check Quality

Compare LoRA, QLoRA and full fine-tuning with worked 7B and 70B memory budgets, published quality results, GPU choices and live prices to estimate run costs.

11 min read
hardware / analysis

Next NVIDIA GPU Roadmap: Plan for Annual Upgrades

Track NVIDIA's data center GPU roadmap through Feynman, see how Rubin Ultra and Kyber plans changed, and choose rental terms around the annual cadence.

11 min read
NVIDIA Groq LPX and Rubin CPX: What Changed for InferenceNVIDIA Vera CPU Replaces Grace and Goes StandaloneNVIDIA Vera Rubin Explained: Specs, Release Date and PriceRubin vs Blackwell: Rent Now, Keep an Exit ClauseTensor vs Pipeline vs Data Parallelism: How to Split a ModelvLLM vs SGLang vs TensorRT-LLM: Choose by WorkloadAI Inference vs Training: What Each Needs From a GPUAMD Instinct MI300X, MI325X, MI355X: Which to RentAWS Trainium vs NVIDIA: Switch Only After a PilotDGX Spark vs. Cloud GPUs: Buy for Memory, Rent FirstGB200 and GB300 NVL72 Explained: Who Needs a Full RackHBM vs GDDR: Pay for Bandwidth After the Model FitsUse nvidia-smi to Check a Cloud GPU Before TrainingInfiniBand vs RoCE Ethernet: Choose by WorkloadNVIDIA DGX Explained: Lineup, Prices and the HGX DifferenceRTX PRO 6000 Blackwell Price: Buy It or Rent It by the HourROCm vs CUDA: What Runs on AMD GPUs and What Needs PortingRTX 6000 vs A6000 vs 6000 Ada vs PRO 6000: Which Is Which
Before you rent

Cloud GPU pricing questions.

What the prices mean, how availability is checked, and what to compare before you commit.

Our data and methodology
How much does a cloud GPU cost per hour?

Among the 15 GPUs in the market summary above, secure on-demand cloud GPUs currently start at $0.44 per GPU-hour (RTX A6000 on LeaderGPU) and the cheapest current prices run up to B300 SXM6 at $7.89/hr on RunPod. Examples: RTX 4090 at $0.45/hr on Vast.ai; A100 SXM4 80GB at $0.80/hr on Vast.ai; H100 SXM5 at $2.90/hr on Ori. All figures are the cheapest current secure on-demand price per GPU-hour tracked by GPUPerHour, excluding spot and peer-to-peer marketplace listings, and are checked periodically. The price depends on the GPU model, VRAM, provider and region; spot instances have separate pricing and can be reclaimed at short notice.

What is the cheapest cloud GPU?

Among the 15 GPUs in GPUPerHour's market summary, the cheapest current secure on-demand GPU is the RTX A6000 at $0.44 per GPU-hour on LeaderGPU. The cheapest GPU for a given job depends on how much VRAM it needs: for example, RTX 4090 at $0.45/hr on Vast.ai and H100 SXM5 at $2.90/hr on Ori are the current floors for those models. All figures are the cheapest current secure on-demand price per GPU-hour tracked by GPUPerHour, excluding spot and peer-to-peer marketplace listings, and are checked periodically. Marketplace and spot listings use different terms and are excluded from this comparison.

How much does an H100 cost per hour?

The cheapest current H100 SXM is $2.90 per GPU-hour on Ori, and 5 providers currently offer secure on-demand H100 SXM capacity. The cheapest current H100 PCIe is $2.59 per GPU-hour on Vast.ai. The median of daily H100 SXM catalog averages in the 30-day window is $2.18 per GPU-hour. All figures are the cheapest current secure on-demand price per GPU-hour tracked by GPUPerHour, excluding spot and peer-to-peer marketplace listings, and are checked periodically. Every current H100 offer, including multi-GPU and NVLink configurations, is listed on the H100 SXM and H100 PCIe pages under /rent.

Which cloud GPU provider is cheapest?

No single provider is cheapest for every GPU. Across the 15 GPUs in the market summary, the cheapest current secure on-demand price is held by Vast.ai (cheapest on 5 of 15), RunPod (cheapest on 3 of 15), Ori (cheapest on 2 of 15). The cheapest provider changes as inventory moves, so GPUPerHour checks 31 providers periodically rather than publishing a static ranking. The provider directory lists each provider's current offer count and price range.

What is a cloud GPU server?

A cloud GPU server is a remote machine with one or more GPUs attached that is rented over the network and billed by the hour (or by the second on some providers) instead of being bought and maintained. Providers offer it as shared virtual machines, dedicated single-tenant instances or bare-metal servers, and it is accessed over SSH, a web terminal or an API, usually with CUDA or ROCm drivers and machine learning frameworks preinstalled. GPUPerHour lists 19,732 such configurations from 31 providers with their per-GPU hourly price and stock status.

How are GPU prices and availability checked?

GPUPerHour collects pricing and stock observations from 31 providers. The current-price summary requires secure on-demand listings, a provider that reports stock, and an observation within 15 minutes of the summary's last refresh; the summary is refreshed at most every 5 minutes. Prices are per GPU-hour; offer counts are configurations, not physical inventory. Cached pages and provider updates can lag changes, so confirm capacity and the final bill with the provider. Historical catalog averages include all pricing types and availability states and cannot measure savings against current quotes.