🖥️ GPU Cost Calculator

Compare cloud GPU rental costs across AWS, GCP, Azure, Lambda Labs, RunPod, CoreWeave, and Vast.ai. Estimate hourly, daily, and monthly spend for H100, A100, L40S, RTX 4090, and T4 GPUs, and instantly spot the cheapest provider.

🖥️ GPU & Provider
On-demand list pricing, per GPU, per hour (2026 estimates). Spot/preemptible pricing can be 30-70% lower.
24 hrs/day
📈 Cost Estimate
Estimated Monthly Cost
Hourly Cost
Daily Cost
Total Hours / Month
Annual Cost
GPUs Selected
Cheapest Provider (this GPU)

📊 Provider Comparison (same GPU type)

Monthly Cost by Provider
⚠️ Prices are 2026 on-demand/list-rate estimates and change frequently — verify current rates on each provider's pricing page before budgeting. Spot, reserved, and committed-use pricing can differ substantially from on-demand rates shown here. Excludes storage, networking, and egress costs.
📊

Enter your details and click Calculate to see results

Guide

About the GPU Cost Calculator

Last updated: August 2026 · Reviewed by the NeftCal editorial team

Renting GPUs in the cloud is the backbone of most AI training and inference budgets, but pricing varies enormously depending on where you rent from — this GPU cost calculator exists because the same NVIDIA H100 80GB card can cost $12.29/hr on AWS or as little as $2.79/hr on RunPod, more than a 4x difference for identical hardware. Pick a GPU model and provider, set your usage pattern, and see hourly, daily, monthly, and annual costs, while comparing every other provider offering that same GPU so you can spot the cheapest option immediately.

What This Calculator Estimates

It multiplies a chosen GPU/provider's on-demand hourly rate by your GPU count and total usage hours to project cost, then re-runs that same math against every other listed provider offering the identical GPU model — surfacing the cheapest alternative for an apples-to-apples comparison rather than a single isolated quote.

Who Should Use This Calculator

ML engineers scoping a training run, startup founders budgeting inference infrastructure, researchers comparing cloud GPU vendors before a grant-funded project, and DevOps/platform teams deciding between a hyperscaler and a GPU-specialized cloud all benefit from seeing the true monthly cost side by side before committing a budget line.

Why GPU Cost Comparison Matters

GPU compute is often the single largest line item in an AI infrastructure budget, whether you're fine-tuning a model, running batch inference, or serving a production LLM endpoint. Hyperscalers (AWS, GCP, Azure) charge a premium for enterprise SLAs, compliance certifications, and integrated networking, while GPU-specialized clouds (RunPod, Lambda Labs, Vast.ai, CoreWeave) often undercut them significantly for raw compute. Knowing the true monthly cost — and the cheapest alternative — before committing to a provider can save thousands of dollars a month at scale.

Real-World Applications

  • Comparing H100/A100 rental rates across AWS, GCP, Azure, RunPod, Lambda Labs, CoreWeave, and Vast.ai before a training run
  • Budgeting a production inference endpoint's monthly GPU spend
  • Deciding between a hyperscaler's enterprise SLA and a GPU cloud's lower price
  • Sizing a GPU choice against a model's memory needs using the VRAM Requirement Calculator first
  • Pairing with the Inference Latency Estimator to weigh cost against response speed

Tips for Accurate Results

  • On-demand list prices shown here are a ceiling — reserved instances, committed-use discounts, and spot/preemptible pricing can lower your effective rate by 30-70%
  • Community cloud and marketplace pricing (RunPod Community Cloud, Vast.ai) fluctuates with spare-capacity supply and demand — treat these figures as representative, not fixed
  • Factor in idle time — if your GPU isn't running 24/7, reducing "hours per day" gives a far more realistic monthly figure than assuming continuous use
  • Remember this tool prices GPU compute only — add storage, egress, and CPU/host costs separately using the Cloud Storage and Data Transfer calculators
  • Check current spot availability and interruption rates before committing a production workload to the cheapest listed option
Formula

The GPU Cost Formula, Explained

How this GPU cost calculator turns an hourly rate and usage pattern into a monthly projection

Cost Formulas
Hourly Cost = Rate per GPU × Number of GPUs

Total Hours = Hours per Day × Days per Month

Monthly Cost = Rate per GPU × Number of GPUs × Total Hours

Annual Cost = Monthly Cost × (365 ÷ Days per Month)

Annualizing by 365 ÷ days-per-month (rather than a flat ×12) correctly scales a partial-month figure — say, a 10-day sprint — up to a realistic full-year run rate instead of understating it.

🏷️

On-Demand vs Spot

On-demand pricing guarantees availability at a fixed hourly rate. Spot/preemptible instances offer 30-70% discounts but can be reclaimed with little notice — better suited to fault-tolerant training jobs than production inference.

☁️

Hyperscaler vs GPU Cloud

AWS, GCP, and Azure bundle GPU instances with enterprise networking, compliance, and support at a premium. Specialized clouds like RunPod, Lambda Labs, and CoreWeave focus purely on GPU compute and typically cost less per hour.

💾

Memory Matters

GPU choice should match your model's VRAM requirement, not just price. Running a model that needs 60GB on a 24GB card simply isn't possible — check the VRAM Calculator first.

⚙️ Why This Formula Works

GPU rental billing is purely usage-based — you pay for exactly the hours a GPU is allocated to you, at a fixed hourly rate per card. Multiplying rate × GPU count × hours captures that linear billing model exactly, which is why comparing providers reduces to comparing a single number: the hourly rate for the same GPU class.

🎯 When to Use It

  • Before committing a training or fine-tuning job to a specific cloud provider
  • When budgeting a production inference endpoint's monthly GPU line item
  • When deciding whether a hyperscaler's premium is worth its SLA and support

📋 Assumptions

  • Usage pattern (hours/day, days/month) is roughly consistent
  • Pricing reflects each provider's standard on-demand/community list rate
  • All GPUs run the full period entered, with no partial-hour billing quirks

⚠️ Limitations of the Formula

  • Doesn't model spot/preemptible discounts or reserved/committed-use pricing
  • Excludes storage, egress, and host CPU/RAM costs — GPU compute only
  • List prices can lag a provider's most recent rate change
  • Doesn't account for interruption risk on community/spot capacity
Walkthrough

Step-by-Step: How to Use the GPU Cost Calculator

From picking a GPU to spotting the cheapest provider

Choose a GPU type and provider

Select a GPU model and provider combination from the dropdown — NVIDIA H100 80GB, A100 80GB, A100 40GB, L40S, RTX 4090, or T4 across AWS, GCP, Azure, Lambda Labs, RunPod, CoreWeave, or Vast.ai — each option shows its on-demand hourly rate.

Enter number of GPUs

Set how many GPUs of that type you need running in parallel, from a single card up to a large multi-node cluster.

Set hours used per day

Enter how many hours per day the GPUs will actually run — continuous 24 hours for training, or fewer hours for intermittent or business-hours-only inference workloads.

Set days per month

Enter how many days per month you expect to run the workload so partial-month or seasonal usage is reflected in the projection.

Calculate and compare providers

Click "Calculate GPU Cost" to see hourly, daily, monthly, and annual totals, plus a side-by-side comparison chart of every other provider offering the same GPU model, with the cheapest option highlighted.

Example

Worked Example

Using the calculator's own default scenario — 1× H100 80GB on RunPod, 24 hrs/day, 30 days

Scenario

Suppose you rent a single NVIDIA H100 80GB on RunPod ($2.79/hr), running continuously (24 hrs/day) for a 30-day month.

Rate$2.79/hr
GPUs1
Usage24 hrs/day × 30 days
Step 1 — Hourly cost: $2.79 × 1 GPU = $2.79/hr.
Step 2 — Total hours: 24 × 30 = 720 hours.
Step 3 — Daily cost: $2.79 × 24 = $66.96/day.
Step 4 — Monthly cost: $2.79 × 1 × 720 = $2,008.80.
Step 5 — Annual cost: $2,008.80 × (365 ÷ 30) = $24,440.40.
Step 6 — Compare to AWS: the same H100 on AWS at $12.29/hr for 720 hours costs $8,848.80/month — about 4.4× RunPod's price for identical hardware.
Monthly Cost (RunPod)
$2,008.80
Annual Cost
$24,440.40
Monthly Cost (AWS)
$8,848.80

Explanation: At this usage level, choosing RunPod over AWS for the identical H100 80GB card saves roughly $6,840 per month — over $82,000 per year. That gap is exactly what this calculator is built to surface: hardware-identical comparisons across the on-demand market so a provider choice is a deliberate decision, not a default.

Interpretation

Understanding Your GPU Cost Result

What your projected monthly GPU spend generally implies

Monthly Cost RangeWhat It Generally MeansRecommended Next Step
Under $500Hobbyist or small experiment scaleA single consumer GPU (RTX 4090) or community-cloud rental fits comfortably
$500 – $3,000Serious side project or small production workloadCompare GPU-specialized clouds against hyperscalers before committing
$3,000 – $15,000Meaningful production training or inference spendEvaluate reserved/committed-use pricing and spot for fault-tolerant jobs
$15,000 – $50,000Scaled multi-GPU workloadNegotiate volume pricing directly with your chosen provider
Over $50,000Large-scale training or fleet-wide inferenceConsider a dedicated enterprise agreement or owning hardware outright

If the cheapest provider differs from your selection: the "Cheapest Provider" result shows exactly how much switching would save for identical hardware — but factor in migration effort, data residency requirements, and support needs before moving a live workload.

If your annual cost looks unexpectedly large: check the "Days Per Month" field — a low value (say, a 10-day sprint) gets annualized up to a full-year run rate, which is intentional but can surprise users expecting a simple ×12.

These are on-demand list-price estimates, not a live quote. Always confirm current rates on the provider's own pricing page before finalizing a budget.

ℹ️

This calculator provides planning estimates only. Actual charges depend on your provider's live pricing, region, committed-use discounts, and any negotiated enterprise rate on your account.

Use Cases

Practical Use Cases for the GPU Cost Calculator

Where comparing GPU rental cost up front genuinely helps

🏋️

Fine-tuning a model

Estimate the total cost of a multi-day or multi-week fine-tuning job before starting it.

🚀

Serving production inference

Project the monthly GPU line item for a live inference endpoint under continuous load.

Choosing spot vs on-demand

Compare on-demand cost against a discounted spot/community rate to decide if the risk is worth it.

🎓

Training a research model

Budget a grant-funded or academic training run across multiple candidate providers.

⚖️

Comparing cloud GPU providers

Run the same hardware and usage pattern across AWS, GCP, Azure, RunPod, and Lambda Labs side by side.

🏢

Enterprise vs GPU-cloud tradeoff

Weigh a hyperscaler's SLA and compliance certifications against a GPU cloud's lower hourly rate.

🧪

Prototyping before scale-up

Test a small GPU configuration's cost before committing to a larger multi-node cluster.

📊

Startup burn-rate planning

Fold projected GPU spend into a startup's monthly infrastructure burn-rate model.

🏗️

Build-vs-rent decisions

Use the monthly and annual figures as the "rent" side of a buy-vs-rent hardware comparison.

🌍

Multi-region deployment costing

Check whether GPU pricing differs meaningfully by provider region before choosing a deployment zone.

📦

Batch job scheduling

Estimate cost for intermittent batch training jobs that don't need 24/7 GPU allocation.

🔁

Re-checking cost after a price change

Re-run the numbers whenever a provider publishes a new on-demand rate.

Pros & Cons

Benefits and Limitations

What this GPU cost calculator does well, and where it can't replace a live quote

✅ Benefits

  • Free, instant, and requires no signup or account
  • Covers 7 major providers across 6 GPU classes in one place
  • Automatically finds and highlights the cheapest provider for the same GPU
  • Side-by-side provider comparison chart at your exact usage level
  • Rolls a single hourly rate into daily, monthly, and annual projections
  • Accounts for partial-month usage via proper annualization
  • Covers both hyperscalers and specialized GPU clouds
  • Downloadable plain-text summary of your estimate
  • Fast-loading, mobile-friendly, runs entirely in your browser
  • Useful as a repeatable check whenever provider pricing changes
  • Helps quantify the real dollar gap between providers, not just relative claims
  • Pairs naturally with the VRAM Calculator for a full sizing-to-cost workflow

⚠️ Limitations

  • Uses on-demand/community list pricing, not live spot auction rates
  • Doesn't model reserved instance or committed-use discounts
  • Excludes storage, egress, and host CPU/RAM costs
  • Doesn't account for interruption risk on spot/community capacity
  • Pricing snapshots can lag a provider's most recent rate change
  • Doesn't model regional price variation within a single provider
  • Not a substitute for your provider's live billing dashboard
  • Doesn't factor in data egress costs from switching providers
Reference

GPU Class & Provider Pricing Comparison

Approximate 2026 on-demand hourly rates by GPU class

GPU ClassVRAMCheapest ProviderHyperscaler Rate
H100 80GB80GBRunPod — $2.79/hrAWS — $12.29/hr
A100 80GB80GBRunPod — $1.64/hrAWS — $5.12/hr
A100 40GB40GBLambda Labs — $1.29/hrAWS — $2.74/hr
L40S48GBRunPod — $0.86/hrLambda Labs — $1.09/hr
RTX 409024GBVast.ai — $0.35/hrRunPod — $0.44/hr
T416GBGCP — $0.35/hrAWS — $0.526/hr

Common Mistakes and Expert Tips

❌ Common Mistakes

  • Assuming 24/7 usage when the workload actually runs a few hours a day
  • Ignoring storage and egress costs, which aren't included in this GPU-only estimate
  • Comparing sticker price without checking VRAM fits the target model first
  • Treating community/spot pricing as guaranteed rather than variable
  • Forgetting the "Days Per Month" field affects annualization, not just monthly cost
  • Not re-checking pricing after a provider publishes a rate change

💡 Expert Tips & Best Practices

📝

Summary: This GPU cost calculator gives you an instant, free comparison of cloud GPU rental pricing across 7 major providers and 6 GPU classes, automatically surfacing the cheapest option for identical hardware. Pair it with the VRAM Requirement Calculator and Cloud Cost Calculator for a complete AI infrastructure budgeting workflow.

FAQ

Frequently Asked Questions

Common questions about GPU cost calculator estimates

Which cloud provider has the cheapest H100 GPUs?
Among the providers tracked here, RunPod ($2.79/hr) and Lambda Labs ($2.99/hr) offer the cheapest on-demand H100 80GB pricing, well below the hyperscalers — AWS ($12.29/hr), Azure ($12.24/hr), and GCP ($11.06/hr). CoreWeave sits in between at $4.76/hr. Specialized GPU clouds are typically cheaper because they don't bundle the same enterprise networking, compliance, and support tiers as the hyperscalers.
Why are hyperscaler GPU prices so much higher than GPU clouds like RunPod or Lambda?
AWS, GCP, and Azure price GPU instances as part of a broader enterprise platform that includes SLAs, dedicated support, compliance certifications, VPC networking, and integration with dozens of other managed services. GPU-specialized clouds like RunPod, Lambda Labs, and Vast.ai focus narrowly on GPU compute, run leaner operations, and in Vast.ai's case broker spare capacity from independent hosts, all of which lowers the price per hour.
Should I use spot/community pricing instead of on-demand?
Spot and community-cloud instances (like RunPod Community Cloud or Vast.ai) can cost 30-70% less than on-demand, but they can be preempted or reclaimed with little notice. They're a good fit for fault-tolerant batch training jobs with checkpointing, but risky for latency-sensitive production inference. This calculator uses on-demand/community list pricing, not spot auctions, so treat the numbers as a ceiling, not a guarantee.
How much VRAM do I need before choosing a GPU here?
That depends on your model size and quantization. Use our VRAM Calculator to estimate the VRAM required for a specific model (e.g., Llama 3 70B at FP16 or INT4), then come back here to compare hourly costs for GPUs with enough memory — H100/A100 80GB for large models, L40S or RTX 4090 for smaller ones.
Does this calculator include storage, networking, or egress costs?
No — this tool estimates only the GPU compute rental cost (hourly rate × number of GPUs × hours used). Storage, data egress, and CPU/RAM for the host instance are typically billed separately. Use our Cloud Storage Cost Calculator and Data Transfer Cost Calculator for those components.
How is the monthly GPU cost calculated?
Monthly cost equals the hourly rate for your chosen GPU/provider, multiplied by the number of GPUs, multiplied by total hours (hours per day × days per month). For example, one RunPod H100 at $2.79/hr run 24 hours a day for 30 days equals 720 total hours, or $2,008.80 for the month. Annual cost scales that monthly figure by 365 ÷ days per month to normalize for partial months.
What's the difference between A100 80GB and A100 40GB pricing?
Both use the same A100 compute die, but the 80GB variant costs noticeably more per hour — on AWS, $5.12/hr for 80GB versus $2.74/hr for 40GB — because it doubles usable VRAM for larger models or bigger batch sizes. Choose 40GB only if your model and batch size comfortably fit; otherwise you risk out-of-memory errors mid-training or mid-inference.
Is the RTX 4090 a good choice for AI workloads?
The RTX 4090 is a strong budget option for inference, fine-tuning smaller models, and experimentation — at $0.35–$0.44/hr on community clouds it's a fraction of an H100's cost. Its 24GB of VRAM limits it to smaller models or heavily quantized large ones, and it lacks the multi-GPU NVLink interconnect and enterprise reliability of data-center cards, so it's better suited to solo developers and researchers than production training clusters.
How much does it cost to train a model on 8 H100s for a week?
Using RunPod's $2.79/hr H100 rate: 8 GPUs × $2.79 = $22.32/hr, × 24 hours × 7 days = 168 total hours, giving roughly $3,749.76 for the week. The same run on AWS at $12.29/hr per GPU would cost about $16,517.76 — over four times as much for identical hardware, which is exactly the kind of gap this calculator is built to surface.
Why does the "Days Per Month" field affect my annual cost?
The calculator first computes your monthly cost from the days and hours you enter, then annualizes it as monthly cost × (365 ÷ days per month) rather than simply multiplying by 12. This correctly scales a partial-month figure (say, a 10-day sprint) up to a full-year run-rate instead of understating it, and keeps a full 30- or 31-day month close to a standard ×12 estimate.
What does "Cheapest Provider" mean in the results?
After you calculate, the tool scans every provider offering the exact same GPU model you selected (for example, all six H100 80GB listings) and highlights whichever has the lowest hourly rate, marking your current selection with a "current" badge if it's already the cheapest. This makes it immediate to see whether switching providers for the same hardware would save money.
Are these prices per GPU or per instance?
Prices shown are per individual GPU, per hour. If a cloud instance bundles multiple GPUs (common on 8-GPU H100 nodes), the calculator's "Number of GPUs" field lets you multiply the per-GPU rate by however many cards you're renting, whether that's a single card or a full multi-GPU node.
How often is GPU pricing on this page updated?
The NeftCal editorial team reviews on-demand GPU pricing periodically and updates the rates shown whenever a major provider publishes a rate change, with a full audit at least quarterly. GPU cloud pricing — especially on community platforms like Vast.ai and RunPod — can shift with supply and demand, so always confirm the live rate on the provider's own pricing page before committing budget.
What's the cheapest way to run a small inference workload?
For a small model that fits comfortably in 16-24GB of VRAM, an RTX 4090 on RunPod ($0.44/hr) or Vast.ai ($0.35/hr) is typically the cheapest option, especially if you only need a few hours a day rather than 24/7 uptime. For larger models needing more VRAM headroom or higher reliability, an L40S ($0.86–$1.09/hr) is a reasonable step up before jumping to A100-class pricing.
Can I use this calculator for on-premise GPU cost comparison?
Not directly — this tool models cloud rental rates, not the upfront capital cost, depreciation, power, and cooling of owning hardware. As a rough rule of thumb, renting is usually cheaper for short-term or bursty workloads, while owning can pay off if you'll run a GPU near-continuously for a year or more. Use the monthly and annual figures here as the "rent" side of a build-vs-rent comparison.
How do I decide between L40S and A100 for inference?
The L40S (48GB VRAM, ~$0.86-$1.09/hr) costs roughly half of an A100 80GB (~$1.64-$5.12/hr) and is well-suited to inference workloads that don't need the A100's higher memory bandwidth or multi-instance GPU (MIG) partitioning. Choose A100 when serving very large models, running high-throughput batched inference, or needing guaranteed data-center-grade reliability; choose L40S when your model fits comfortably in 48GB and cost efficiency matters more than peak throughput.
Learn More

Authoritative Resources on GPU Cloud Pricing

Official documentation to complement this calculator — always verify live rates before finalizing a budget

Related Calculators

Explore other AI & infrastructure tools