{"id":138740,"date":"2026-10-06T14:17:04","date_gmt":"2026-10-06T14:17:04","guid":{"rendered":"\/uk\/tutorials\/cloud-gpu-pricing"},"modified":"2026-10-06T14:17:04","modified_gmt":"2026-10-06T14:17:04","slug":"cloud-gpu-pricing","status":"publish","type":"post","link":"\/uk\/tutorials\/cloud-gpu-pricing\/","title":{"rendered":"Cloud GPU pricing compared: what GPUs cost per hour in 2026"},"content":{"rendered":"

Cloud GPU pricing ranges from under $0.50 per hour for an entry-level card to over $12 for the most powerful ones. For the same GPU, the price can still vary by as much as sevenfold depending on which provider you rent from. That gap comes from how providers package the hardware, not from the chip itself.<\/p>

What you pay depends heavily on how that access is structured: some providers rent you just the GPU; others bundle it inside a full server with dozens of processors and terabytes of memory, charge for everything, and require you to rent several GPUs at once.<\/p>

The pricing model adds a second layer: on-demand, interruptible, or reserved access can halve or double the rate for the same card.<\/p>

The advertised rate also doesn’t capture the full cost: storage fees, data transfer charges, and idle GPUs running between jobs can push a real monthly bill well above what the hourly rate suggests.<\/p>

How much do cloud GPUs cost per hour?<\/h2>

Cloud GPUs cost between $0.14 and $12.29 per GPU-hour (the cost of renting one GPU for one hour) at on-demand (pay-as-you-go) list prices, as of September 2026.<\/p>

The comparison table covers seven GPU models from seven providers across US regions. The VRAM column shows each GPU’s on-chip memory, which determines how large a model the GPU can run.<\/p>

GPU<\/strong><\/p><\/td>

VRAM<\/strong><\/p><\/td>

Hostinger<\/strong><\/p><\/td>

RunPod<\/strong><\/p><\/td>

Vast.ai<\/strong><\/p><\/td>

Lambda<\/strong><\/p><\/td>

AWS<\/strong><\/p><\/td>

GCP<\/strong><\/p><\/td>

Azure<\/strong><\/p><\/td><\/tr>

RTX 4090<\/span><\/p><\/td>

24 GB<\/span><\/p><\/td>

$0.38<\/strong><\/p><\/td>

$0.74<\/strong><\/p><\/td>

from <\/span>$0.14<\/strong><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td><\/tr>

RTX A6000<\/span><\/p><\/td>

48 GB<\/span><\/p><\/td>

–<\/span><\/p><\/td>

$0.53<\/strong><\/p><\/td>

from <\/span>$0.28<\/strong><\/p><\/td>

$1.09<\/strong><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td><\/tr>

RTX PRO 6000<\/span><\/p><\/td>

96 GB<\/span><\/p><\/td>

$0.60<\/strong><\/p><\/td>

$2.09<\/strong><\/p><\/td>

from <\/span>$0.89<\/strong><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

$4.50<\/strong>†<\/span><\/p><\/td>

–<\/span><\/p><\/td><\/tr>

L40S<\/span><\/p><\/td>

48 GB<\/span><\/p><\/td>

$0.92*<\/strong><\/p><\/td>

$0.99<\/strong><\/p><\/td>

from <\/span>$0.47<\/strong><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td>

–<\/span><\/p><\/td><\/tr>

A100 80GB<\/span><\/p><\/td>

80 GB<\/span><\/p><\/td>

$1.43<\/strong><\/p><\/td>

$1.39<\/strong><\/p><\/td>

from <\/span>$0.44<\/strong><\/p><\/td>

from <\/span>$1.99<\/strong>§<\/span><\/p><\/td>

$3.43†<\/span><\/p><\/td>

$3.67<\/strong>§<\/span><\/p><\/td>

$3.67<\/strong>†<\/span><\/p><\/td><\/tr>

H100<\/span><\/p><\/td>

80 GB<\/span><\/p><\/td>

–<\/span><\/p><\/td>

$2.89<\/strong><\/p><\/td>

from <\/span>$1.33<\/strong><\/p><\/td>

$3.29<\/strong><\/p><\/td>

$6.88<\/span><\/p><\/td>

$11.06<\/strong>†<\/span><\/p><\/td>

$12.29<\/strong>†<\/span><\/p><\/td><\/tr>

B200<\/span><\/p><\/td>

192 GB<\/span><\/p><\/td>

$4.50<\/strong><\/p><\/td>

$6.79<\/strong><\/p><\/td>

from <\/span>$4.38<\/strong><\/p><\/td>

from <\/span>$6.69<\/strong><\/p><\/td>

–<\/span><\/p><\/td>

$16.11<\/strong>‡<\/span><\/p><\/td>

–<\/span><\/p><\/td><\/tr><\/tbody><\/table><\/figure>

On-demand rates, normalized per GPU-hour, US regions, as of September 2026. RunPod: Secure Cloud tier. Vast.ai: marketplace range; prices change daily. (–) = not offered.<\/em><\/p>

† Whole-VM or minimum 8-GPU node; per-GPU rate derived from instance total.<\/em><\/p>

‡ Reservation required; no standard on-demand.<\/em><\/p>

* Availability varies.<\/em><\/p>

§ Lambda $1.99 and GCP $3.67 reflect A100 40GB variants (cheapest option). Lambda 80GB SXM is approximately $2.79; GCP a2-ultragpu 80GB is approximately $5.03.<\/em><\/p>

Beyond the per-GPU rate, the tier you pick – shared community hardware or dedicated data-center capacity –, regional availability, and uptime commitments all affect the real cost of a workload. On those dimensions, <\/a>comparing GPU cloud providers<\/a> often changes the final choice more than a small rate difference would.<\/p>

\n \n \n \n <\/svg>\n Important\n <\/span>\n

\n GPU cloud pricing updates monthly and sometimes weekly; marketplace platforms like Vast.ai can change hourly. The rates above were verified in September 2026 – re-check primary pricing pages before you deploy anything. \n <\/p><\/div><\/p>

How does cloud GPU pricing work?<\/h2>

Cloud GPU pricing runs on three models: on-demand pay-as-you-go, interruptible spot capacity at a significant discount, and reserved terms for predictable long-running workloads.<\/p>

On-demand<\/h3>

On-demand gives you immediate access at a fixed rate with no commitment. You pay for the whole time the instance (your running virtual server) is running, whether the GPU is processing a job or sitting idle. It’s the default model for variable or short-term workloads and the basis for every rate in the comparison table.<\/p>

Spot and interruptible<\/h3>

Spot pricing cuts costs from 50 to 91 percent depending on the provider, but the instance can be reclaimed with little notice when the provider needs the capacity back. Google Cloud Platform (GCP) gives 30 seconds of warning; Amazon Web Services (AWS) gives two minutes.<\/p>

Spot only makes sense with proper checkpointing: saving your job’s progress at intervals so it can resume from the last save point after an interruption, rather than from scratch. A job with no checkpoints that gets evicted mid-run can end up costing more than an on-demand run would have.<\/p>

\n
\n \n \n <\/svg>\n

\n Pro tip <\/p>\n <\/div>\n

Spot prices are market rates, not guaranteed discounts. On bid-based platforms, a supply crunch can push the spot rate close to or above on-demand. Always check the live spot rate before assuming it's cheaper than on-demand. <\/p>\n <\/div>\n <\/p>

Provider<\/strong><\/p><\/td>

Spot\/interruptible model<\/strong><\/p><\/td>

Typical discount<\/strong><\/p><\/td>

Notes<\/strong><\/p><\/td><\/tr>

AWS<\/span><\/p><\/td>

Spot Instances<\/span><\/p><\/td>

~57%<\/span><\/p><\/td>

~$2.96\/GPU-hr for H100 (p5.48xlarge); fluctuates<\/span><\/p><\/td><\/tr>

GCP<\/span><\/p><\/td>

Spot VMs<\/span><\/p><\/td>

60–91%<\/span><\/p><\/td>

30-second eviction notice<\/span><\/p><\/td><\/tr>

Azure<\/span><\/p><\/td>

Spot VMs<\/span><\/p><\/td>

~80%<\/span><\/p><\/td>

30-second eviction notice<\/span><\/p><\/td><\/tr>

RunPod<\/span><\/p><\/td>

Community Cloud<\/span><\/p><\/td>

~50%<\/span><\/p><\/td>

No published spot rate card<\/span><\/p><\/td><\/tr>

Vast.ai<\/span><\/p><\/td>

Interruptible<\/span><\/p><\/td>

50%+<\/span><\/p><\/td>

Host-set; varies daily<\/span><\/p><\/td><\/tr><\/tbody><\/table><\/figure>

Reserved and committed<\/h3>

Reserved pricing locks in a GPU for a fixed term in exchange for a lower effective hourly rate. Discounts range from 25 to 65 percent, depending on the provider and commitment length.<\/p>

Provider<\/strong><\/p><\/td>

Discount<\/strong><\/p><\/td>

Term options<\/strong><\/p><\/td><\/tr>

AWS<\/span><\/p><\/td>

25–45%<\/span><\/p><\/td>

Capacity Blocks (variable), Savings Plans<\/span><\/p><\/td><\/tr>

GCP<\/span><\/p><\/td>

up to 65%<\/span><\/p><\/td>

1 or 3 years (Committed Use Discounts)<\/span><\/p><\/td><\/tr>

Azure<\/span><\/p><\/td>

up to 63%<\/span><\/p><\/td>

1 or 3 years; 5 years on ND H100 v5 (Reserved Instances)<\/span><\/p><\/td><\/tr>

Vast.ai<\/span><\/p><\/td>

Up to 50%<\/span><\/p><\/td>

1, 3, or 6 months<\/span><\/p><\/td><\/tr>

Lambda<\/span><\/p><\/td>

1-Click Clusters from $5.54\/GPU\/hr<\/span><\/p><\/td>

2 weeks – 1 year; 16+ GPUs minimum; 1yr+ contact sales<\/span><\/p><\/td><\/tr>

RunPod<\/span><\/p><\/td>

Contact sales<\/span><\/p><\/td>

Enterprise agreements<\/span><\/p><\/td><\/tr><\/tbody><\/table><\/figure>

Billing granularity<\/h3>

Providers charge in different units, and the difference matters most for short or interrupted jobs. RunPod and Vast.ai bill per second; Lambda bills per-minute; AWS, GCP, and Azure bill per second with a one-minute minimum. A 16-minute job on RunPod costs roughly 27 percent of the hourly rate; on Lambda it costs the full hour.<\/p>

Hostinger bills hourly using a prepaid credit system (1 credit = $0.01), with the first full hour deducted at the moment of deployment. Destroying an instance mid-hour does not return unused time.<\/p>

Cloud GPU pricing by model<\/h2>

Cloud GPU models fall into several pricing tiers, from under $0.14\/hr for the RTX 4090 to $6.79\/hr for the B200 on standard on-demand – and higher still at hyperscalers, where a reserved B200 runs $16.11\/hr. Each tier offers different VRAM capacity and suits different jobs.<\/p>

RTX 4090 pricing per hour<\/h3>

The RTX 4090 costs from $0.14<\/strong> per hour on Vast.ai and $0.38<\/strong> per hour on Hostinger, as of September 2026. RunPod lists it at $0.34<\/strong> on Community Cloud and $0.74<\/strong> on Secure Cloud.<\/p>

Community Cloud uses hardware contributed by independent hosts that RunPod has screened; Secure Cloud uses data-center-grade infrastructure with higher reliability guarantees.<\/p>

No hyperscaler carries the RTX 4090, since GPUs originally built for gaming and workstation use aren’t available on AWS, GCP, or Azure.<\/p>

With 24 GB of VRAM, the 4090 handles most 7-billion-parameter models at full 16-bit precision and up to roughly 30 billion parameters at 4-bit quantization, a compression method that reduces memory usage at a small cost to accuracy.<\/p>

Ollama, a free tool for running open-source AI models on cloud or local hardware, is a natural fit for this card. A<\/a>t the 24 GB tier, choosing a GPU for Ollama<\/a> means most popular open-weight models run without additional configuration.<\/p>

The 4090 is also a popular choice for Stable Diffusion (an open-source image generation model) and ComfyUI (a visual workflow tool for running it), where generating images quickly matters more than raw training speed.<\/p>

RTX A6000 and RTX PRO 6000 pricing<\/h3>

The RTX A6000 starts at $0.28<\/strong> per hour on Vast.ai, $0.53<\/strong> on RunPod, and $1.09<\/strong> on Lambda, as of September 2026. Built on NVIDIA’s older Ampere architecture with 48 GB of VRAM, it keeps NVLink, which lets multiple GPUs share memory directly – a feature the newer L40S drops.<\/p>

The RTX PRO 6000 (96 GB, built on NVIDIA’s newer Blackwell architecture) is its successor. Hostinger lists it at $0.60<\/strong> per hour, while RunPod charges $2.09. GCP’s G4 series bundles the same card with 48 vCPUs (virtual CPU cores) and 180 GB of RAM into a single virtual machine from approximately $4.50<\/strong> per hour.<\/p>

The memory difference is the decision point. At 48 GB, the A6000 can hold a model of roughly 20 billion parameters at full FP16 precision (16-bit floating point, the standard format for full-accuracy AI model weights) once you leave room for the context window; larger models require quantization to fit.<\/p>

With 96 GB, the PRO 6000 can handle most 70-billion-parameter models on a single card at 4-bit quantization, without needing to link multiple GPUs.<\/p>

L40S pricing per hour<\/h3>

The L40S costs $0.92<\/strong> per hour on Hostinger and $0.99<\/strong> on RunPod Secure Cloud, as of September 2026. It’s built on NVIDIA’s Ada Lovelace architecture with 48 GB of VRAM and native FP8 support – FP8 is an 8-bit number format that reduces memory usage and can speed up AI inference workloads.<\/p>

That FP8 support makes it the better option for inference (running a trained AI model to generate outputs) in the 48 GB tier compared to the older A6000.<\/p>

GCP’s G2 series offers the L4 (24 GB) from $0.71<\/strong> per hour. The L4 and L40S share the same Ada Lovelace architecture, but the L4 is a low-power inference card with half the VRAM and a fraction of the throughput – it draws 72 W against the L40S’s 350 W. If both appear in GCP’s pricing for an inference workload, the L4 is not a cheaper L40S: it is a much smaller card.<\/p>

\n \n \n \n <\/svg>\n Important\n <\/span>\n

\n L40S availability on Hostinger fluctuates. If the card doesn’t appear in the deploy picker, it’s simply out of stock, not unavailable in your region. Hostinger shows live inventory before you commit to a deploy, so what you see is what’s actually available. \n <\/p><\/div><\/p>

A100 80GB pricing per hour<\/h3>

The A100 80GB starts at $1.39<\/strong> per hour on RunPod and $1.43<\/strong> per hour on Hostinger, rising to $3.67<\/strong>–$5.03<\/strong> per hour at hyperscalers.<\/p>

RunPod’s $1.39 rate is the PCIe variant; the SXM variant (faster for multi-GPU training) is $1.59<\/strong>. At hyperscalers, the gap widens sharply: Azure’s NC24ads A100 v4 is $3.67<\/strong> per hour for a single-GPU VM; the per-GPU rate on AWS’s p4de (8× A100 80GB) works out to roughly $3.43<\/strong>, and GCP’s a2-highgpu-1g (A100 40GB) runs $3.67<\/strong>.<\/p>

The A100’s 80 GB of high-bandwidth memory (HBM2e) handles fine-tuning (adapting an existing model for a specific task) and inference for most models up to 70 billion parameters in quantized form. At the specialist-cloud rate of $1.39–1.43 per hour, it’s the default step up from the 48 GB tier for training runs that need more memory.<\/p>

H100 pricing per hour<\/h3>

An H100 costs $2.89<\/strong> per hour on RunPod and up to $12.29<\/strong> on Azure, as of September 2026. That is a more than 4x spread between the cheapest specialist cloud and the most expensive hyperscaler for the same chip. Hostinger doesn’t offer an H100; the B200 covers its top tier.<\/p>