Cloud GPU pricing compared: what GPUs cost per hour in 2026

Cloud GPU pricing ranges from under $0.50 per hr for an entry-level cardstock to complete $12 for the most mighty ones. For the identical GPU, the cost can motionless change by as much as sevenfold depending on which provider you rent from. That gap comes from how providers bundle the hardware, not from the part itself.

What you pay depends heavily on how that admission is structured: several providers rent you fair the GPU; others bundle it inner a complete server alongside dozens of processors and terabytes of memory, accusation for everything, and necessitate you to rent multiple GPUs at once.

The pricing example adds a second layer: on-demand, interruptible, or reserved admission can halve or twice the charge for the identical card.

The advertised charge additionally doesn’t grasp the complete cost: retention fees, data transfer charges, and idle GPUs operating between jobs can shove a genuine monthly invoice fine complete what the hourly charge suggests.

How much do haze GPUs disbursal per hour?

Cloud GPUs disbursal between $0.14 and $12.29 per GPU-hour (the disbursal of renting one GPU for one hour) at on-demand (pay-as-you-go) catalog prices, as of September 2026.

The difference array covers seven GPU models from seven providers throughout US regions. The VRAM pillar shows all GPU’s on-chip memory, which determines how ample a example the GPU can run.

On-demand rates, normalized per GPU-hour, US regions, as of September 2026. RunPod: Secure Cloud tier. Vast.ai: market range; prices alter daily. (–) = not offered.

† Whole-VM or minimum 8-GPU node; per-GPU charge derived from case total.

‡ Reservation required; no norm on-demand.

* Availability varies.

§ Lambda $1.99 and GCP $3.67 indicate A100 40GB variants (cheapest option). Lambda 80GB SXM is about $2.79; GCP a2-ultragpu 80GB is about $5.03.

Beyond the per-GPU rate, the tier you choice – shared community hardware or dedicated data-center capability –, local availability, and uptime commitments all power the genuine disbursal of a workload. On those dimensions, comparing GPU haze providers frequently changes the final choice additional than a small charge difference would.

How does haze GPU pricing work?

Cloud GPU pricing runs on three models: on-demand pay-as-you-go, interruptible place capability at a important discount, and reserved conditions for predictable long-running workloads.

On-demand

On-demand gives you contiguous admission at a fixed charge alongside no commitment. You pay for the entire period the case (your operating virtual server) is running, whether the GPU is handling a job or sitting idle. It’s the default example for changeable or short-term workloads and the basis for all charge in the difference table.

Spot and interruptible

Spot pricing cuts expenses from 50 to 91 percent depending on the provider, but the case can be reclaimed alongside small notice whenever the provider needs the capability back. Google Cloud Platform (GCP) gives 30 seconds of warning; Amazon Web Services (AWS) gives two minutes.

Spot lone makes awareness alongside appropriate checkpointing: preservation your job’s advancement at intervals so it can resume from the final preserve item following an interruption, fairly than from scratch. A job alongside no checkpoints that gets evicted mid-run can end up costing additional than an on-demand run would have.

Reserved and committed

Reserved pricing locks in a GPU for a fixed term in toggle for a lesser productive hourly rate. Discounts range from 25 to 65 percent, depending on the provider and dedication length.

Billing granularity

Providers accusation in distinct units, and the difference matters most for abbreviated or interrupted jobs. RunPod and Vast.ai invoice per second; Lambda bills per-minute; AWS, GCP, and Azure invoice per second alongside a one-minute minimum. A 16-minute job on RunPod expenses approximately 27 percent of the hourly rate; on Lambda it expenses the complete hour.

Hostinger bills hourly using a prepaid credit scheme (1 credit = $0.01), alongside the archetypal complete hr deducted at the instant of deployment. Destroying an case mid-hour does not come back unused time.

Cloud GPU pricing by model

Cloud GPU models autumn into multiple pricing tiers, from under $0.14/hr for the RTX 4090 to $6.79/hr for the B200 on norm on-demand – and higher motionless at hyperscalers, anywhere a reserved B200 runs $16.11/hr. Each tier offers distinct VRAM capability and suits distinct jobs.

RTX 4090 pricing per hour

The RTX 4090 expenses from $0.14 per hr on Vast.ai and $0.38 per hr on Hostinger, as of September 2026. RunPod lists it at $0.34 on Community Cloud and $0.74 on Secure Cloud.

Community Cloud uses hardware contributed by autonomous hosts that RunPod has screened; Secure Cloud uses data-center-grade infrastructure alongside higher reliability guarantees.

No hyperscaler carries the RTX 4090, since GPUs initially built for gaming and workstation use aren’t accessible on AWS, GCP, or Azure.

With 24 GB of VRAM, the 4090 handles most 7-billion-parameter models at complete 16-bit precision and up to approximately 30 milliard parameters at 4-bit quantization, a compression method that reduces recollection use at a small disbursal to accuracy.

Ollama, a liberated tool for operating open-source AI models on haze or local hardware, is a natural fit for this card. At the 24 GB tier, choosing a GPU for Ollama method most famous open-weight models run without additional configuration.

The 4090 is additionally a famous choice for Stable Diffusion (an open-source depiction generation model) and ComfyUI (a ocular workflow tool for operating it), anywhere generating images quickly matters additional than raw training speed.

RTX A6000 and RTX PRO 6000 pricing

The RTX A6000 starts at $0.28 per hr on Vast.ai, $0.53 on RunPod, and $1.09 on Lambda, as of September 2026. Built on NVIDIA’s older Ampere architecture alongside 48 GB of VRAM, it keeps NVLink, which lets multiple GPUs portion recollection immediately – a characteristic the newer L40S drops.

The RTX PRO 6000 (96 GB, built on NVIDIA’s newer Blackwell architecture) is its successor. Hostinger lists it at $0.60 per hour, during RunPod charges $2.09. GCP’s G4 sequence bundles the identical cardstock alongside 48 vCPUs (virtual CPU cores) and 180 GB of RAM into a sole virtual device from about $4.50 per hour.

The recollection difference is the decision point. At 48 GB, the A6000 can clasp a example of approximately 20 milliard parameters at complete FP16 precision (16-bit floating point, the norm format for full-accuracy AI example weights) formerly you depart area for the environment window; larger models necessitate quantization to fit.

With 96 GB, the PRO 6000 can grip most 70-billion-parameter models on a sole cardstock at 4-bit quantization, without needing to nexus multiple GPUs.

L40S pricing per hour

The L40S expenses $0.92 per hr on Hostinger and $0.99 on RunPod Secure Cloud, as of September 2026. It’s built on NVIDIA’s Ada Lovelace architecture alongside 48 GB of VRAM and native FP8 assistance – FP8 is an 8-bit figure format that reduces recollection use and can speed up AI conclusion workloads.

That FP8 assistance makes it the improved choice for conclusion (running a trained AI example to create outputs) in the 48 GB tier compared to the older A6000.

GCP’s G2 sequence offers the L4 (24 GB) from $0.71 per hour. The L4 and L40S portion the identical Ada Lovelace architecture, but the L4 is a low-power conclusion cardstock alongside fractional the VRAM and a fraction of the throughput – it draws 72 W against the L40S’s 350 W. If the two appear in GCP’s pricing for an conclusion workload, the L4 is not a cheaper L40S: it is a much smaller card.

A100 80GB pricing per hour

The A100 80GB starts at $1.39 per hr on RunPod and $1.43 per hr on Hostinger, rising to $3.67–$5.03 per hr at hyperscalers.

RunPod’s $1.39 charge is the PCIe variant; the SXM type (faster for multi-GPU training) is $1.59. At hyperscalers, the gap widens sharply: Azure’s NC24ads A100 v4 is $3.67 per hr for a single-GPU VM; the per-GPU charge on AWS’s p4de (8× A100 80GB) plant out to approximately $3.43, and GCP’s a2-highgpu-1g (A100 40GB) runs $3.67.

The A100’s 80 GB of high-bandwidth recollection (HBM2e) handles fine-tuning (adapting an existing example for a particular task) and conclusion for most models up to 70 milliard parameters in quantized form. At the specialist-cloud charge of $1.39–1.43 per hour, it’s the default stage up from the 48 GB tier for training runs that need additional memory.

H100 pricing per hour

An H100 expenses $2.89 per hr on RunPod and up to $12.29 on Azure, as of September 2026. That is a additional than 4x dispersed between the cheapest expert haze and the most costly hyperscaler for the identical chip. Hostinger doesn’t recommendation an H100; the B200 covers its top tier.

• RunPod: $2.89 (PCIe); $3.29 (SXM); $3.19 (NVL).
• Lambda: $3.29 (PCIe); from $3.99 (SXM; cheapest per-GPU at the 8-GPU config, sole GPU accessible at $4.29).
• Nebius: $3.85.
• DigitalOcean: $4.41.
• AWS p5.4xlarge: $6.88 (single H100 SXM5; no minimum node).
• Azure NC H100 v5: $6.98 (NVL form factor, sole GPU).
• GCP A3 (a3-highgpu-8g): from $11.06 per GPU (8-GPU minimum on-demand).
• Azure ND H100 v5: $12.29 per GPU ($98.32 for the 8-GPU node).

PCIe, SXM, and NVL are the form factors the H100 ships in. PCIe fits a norm server slot; SXM uses NVIDIA’s proprietary high-power socket alongside NVLink interconnects that let multiple GPUs portion recollection directly, which matters for distributed training (splitting a job throughout multiple GPUs in parallel). NVL pairs two cards and carries additional recollection per GPU.

The SXM premium at expert clouds is approximately 3–10%. For single-card inference, PCIe is the cheaper choice alongside no achievement difference at that scale.

GCP’s A3 on-demand requires a minimum of 8 GPUs per node – if you need two H100s, you motionless pay for eight. Lambda sells H100 SXM in 1×, 2×, 4×, and 8× configurations.

B200 and H200 pricing

The B200 expenses $4.50 per hr on Hostinger, $6.79 on RunPod, from $4.38 on Vast.ai, and from $6.69 on Lambda. The H200 starts at $3.87 on Vast.ai and $4.59 on RunPod. GCP’s A4 sequence starts about $16.11 per GPU-hour but requires a reservation. There’s no norm on-demand option.

The H200 (141 GB HBM3e, a high-bandwidth recollection type) carries approximately 75% additional VRAM than the H100; RunPod lists it at $4.59. GCP’s A3 Ultra and Azure’s ND H200 v5 additionally necessitate reservations and aren’t accessible on norm on-demand.

Both cards sit complete the H100, but in distinct ways: the B200 is a new architecture alongside higher throughput, during the H200 keeps the H100’s compute and adds memory. As these cards rotate into additional extensively available, H100 on-demand rates are apt to keep falling. A100 pricing has already dropped below $1 per hr on several market providers as the market shifts toward newer GPU generations.

Why do hyperscalers accusation additional for the identical GPU?

Hyperscalers (AWS, Google Cloud and Azure) accusation considerably additional than expert GPU clouds for on-demand admission since they rent you a entire virtual machine, not fair the GPU. That bundle bundles compute, memory, and storage, frequently alongside a minimum figure of GPUs per rental, affirmative the compliance certifications and assistance tooling most AI workloads don’t need.

Forced bundling. AWS’s p5.4xlarge pairs one H100 alongside 16 vCPUs (virtual CPU cores) and 256 GB of scheme RAM. Azure’s ND H100 v5 bundles eight GPUs alongside 96 vCPUs and 1.9 TB of RAM. You can’t rent the GPU on its own – the processors and recollection arrive attached in fixed ratios, whether you need them or not.

On Hostinger, the identical B200 GPU is accessible in tiers from $4.50 to $6.00/hr, a 33% dispersed from smallest to largest.

Quota guide time. AWS and GCP the two cap how many high-end GPU instances a new document can initiate – these limits are called quotas, and they default to zero for GPU instances.

Getting AWS P5 admission requires a written justification and typically takes 3–7 endeavor days for approval; GCP’s A3 H100 quota can obtain longer. That delay is a genuine cost: teams blocked on quota frequently end up renting additional capability than they need on any they can admission immediately..

What hyperscalers provision you in return. They run high-speed networking between GPU nodes, up to 3.2 Tbps (terabits per second) on AWS P5 instances, which matters for distributed training throughout dozens of GPUs. Enterprise compliance certifications are built in.

Regional safety spans all important geography. For regulated industries or workloads that need 64+ GPU clusters, the hyperscaler premium buys infrastructure you can’t collect elsewhere.

For developers operating inference, fine-tuning, or depiction generation, most of that bundle is unnecessary. Specialist GPU clouds band it out, and alternatives to RunPod for GPU workloads in the identical tier typically arrive without quota endorsement or minimum node sizes.

The cost gap is most apparent whenever you lone need one GPU, anywhere expert clouds consistently undercut hyperscaler rates.

What you really pay for haze GPUs

The hourly GPU charge is lone part of the cost: storage, data transfer fees, and idle period can shove the genuine invoice up by 20–40%.

Data transfer costs

Moving data out of a provider’s network is liberated on several platforms and expenses up to $0.12 per GB on others, which adds up accelerated on ample workloads. Providers call this egress rate.

The difference scales alongside volume: moving a 140 GB Llama-70B example checkpoint out of AWS expenses about $12.60, during moving a 10 TB training dataset expenses approximately $900. On Hostinger, RunPod Pods, or Lambda, the identical transfers disbursal nothing.

Storage costs

Cloud GPU retention typically runs $0.04–0.20/GB/month, and on most platforms it keeps billing equal following you halt your instance.

• RunPod. $0.10/GB/month during running, $0.20/GB/month whenever stopped. A 200 GB quantity remaining idle for a duration expenses $40.
• Vast.ai. Rates are host-set and change by listing. Storage bills continuously equal whenever stopped, typically at a higher charge than whenever running. Deleting the case is the lone way to halt the charge.
• Lambda. $0.20/GB/month for persistent document storage, which bills equal alongside no GPU case attached.
• AWS. $0.08/GB/month (EBS), billed separately on top of compute.
• GCP. $0.04–0.17/GB/month (standard to SSD persistent disk), billed separately on top of compute.
• Azure. $0.17/GB/month (Premium SSD), billed separately on top of compute.
• Hostinger. Included alongside all case tier. No distinct retention charge.

Idle GPU costs

Idle GPUs invoice at the complete hourly rate, the identical as whenever they are operating a job. A GPU remaining operating following a job finishes expenses exactly as much as one that was computing the entire time.

This is true of all provider: a meeting forgotten overnight expenses a complete night’s compute everywhere. Billing granularity lone changes the rounding at the edges. Hostinger deducts the complete archetypal hr at deploy, so a 10-minute meeting expenses the identical as a 60-minute one. On RunPod or Vast.ai, that identical 10-minute meeting expenses exactly 10 minutes.

How much does Hostinger GPU hosting cost?

Hostinger GPU hosting starts at $0.38 per hr for an RTX 4090 and reaches $7.08 per hr for a dedicated B200, billed hourly in credits alongside no egress fees, as of September 2026.

Every case gets a dedicated NVIDIA GPU – the cardstock isn’t shared between customers – alongside complete base and SSH access. The distinct B200 (Dedicated) tier goes further, giving you the entire underlying server fairly than a portion of it. One-click apps for Ollama, ComfyUI, and Jupyter are accessible at deploy period alongside a basis Ubuntu 24.04 option. Deployment takes a few minutes formerly you choose a GPU and case size.

Each GPU comes in multiple size tiers. The B200’s basis tier (Nano) gives you 1 core, 1 GB RAM, and 10 GB retention at 450 credits per hr ($4.50). The largest shared tier (32 cores, 64 GB RAM, 500 GB storage) expenses 600 credits per hr ($6.00). The dispersed throughout the shared tiers is 33 percent.

Hostinger GPU hosting uses credits purchased in advance. Three group sizes are available:

• 500 credits. $5.00 ($0.0100/credit, basis rate).
• 2,200 credits. $20.00 ($0.0091/credit).
• 6,000 credits. $50.00 ($0.0083/credit).

The complete archetypal hr is deducted at deploy. Destroying an case mid-hour does not come back the unused portion.

Hostinger GPU hosting is US-only (Los Angeles and Dallas, area depending on the GPU) and on-demand only. No place or reserved tiers exist. There is no GPU switching following deploy; changing to a distinct cardstock requires a new case and a data migration.

How to evaluation your monthly haze GPU bill

To evaluation a monthly haze GPU bill, multiply your planned hours by the hourly rate, afterward add retention expenses and egress fees. On hyperscalers the two are billed separately and develop alongside the amount of data you move; on Hostinger they add nothing, during RunPod and Lambda accusation for retention but not egress.

This example uses a sole A100 80GB for LLM (Large Language Model) conclusion using a example specified as Llama or Mistral, 8 hours per day complete 30 days (240 GPU-hours), alongside 10 GB of example weights in retention and 200 GB of data transferred out per month.

At 8 hours per day, Hostinger and RunPod district inside $9 of all other. Azure expenses 2.7 times more, alongside the compute charge driving nearly all the difference.

Applying Hostinger’s $50 credit group (effective $0.0083/credit) reduces the Hostinger total to about $285. At that rate, the gap complete RunPod widens in Hostinger’s favor for this workload.

For manufacturing LLM inference, how you size the GPU to the example matters as much as which provider you pick.

Deploying LLMs on GPU infrastructure alongside vLLM requires matching GPU recollection to example size, which determines whether an A100 or a smaller cardstock is the correct choice for a stated workload.

How to choose the correct haze GPU pricing example for your workload

The correct haze GPU pricing example depends on three factors: how lengthy your jobs run, whether they can be interrupted, and how frequently you need the GPU. For most workloads, that method on-demand admission at a expert cloud.

If your jobs obtain multiple hours and you don’t need to observe them run, place pricing cuts expenses by 50% or more. RunPod’s Spot Pods and Vast.ai’s interruptible tier recommendation the steepest discounts, although the provider can reclaim the GPU mid-job, so preserve your advancement regularly.

If you need a GPU operating continuously for a manufacturing AI assistance alongside accordant person traffic, on-demand pricing stops making sense. Azure’s 3-year A100 booking drops from $3.67 to $1.36 per hour, a 63% decrease for a workload that was going to run anyway. Most teams attain this item whenever they’re serving a live product.

Before picking a provider, inspect the data transfer rate: moving 10 TB out of GCP or AWS can add $800–1,000 per month. The identical transfer is liberated on Hostinger, RunPod Pods, and Lambda. On data-heavy workloads, the transfer fee can outweigh any funds from a cheaper hourly rate.

One method value trying: keep your use and data on a VPS and rotation up a GPU lone whenever you need to run something. You pay for the GPU lone during it runs.

All of the tutorial satisfied on this website is topic to Hostinger's rigorous editorial standards and values.