GPU Cloud Cost Calculator

How Monthly Cost and the Break-Even Are Calculated

Monthly cost = rate per GPU-hour × GPUs × hours a day × days a month. Owning costs electricity: watts ÷ 1,000 × price per kWh. Break-even hours = purchase price ÷ (rental rate − electricity per hour), and months to break even = those hours ÷ your monthly hours.

Worked example with the defaults: 8 hours × 30 days = 240 GPU-hours, × $0.89 = $213.60 a month or $2,563.20 a year. An RTX 4090 is rated at 450 W total graphics power, so it uses 0.45 kWh an hour, about $0.0824 at 18.31¢ per kWh. Each rented hour costs $0.89 − $0.0824 = $0.8076 more than running your own card, and $1,800 ÷ $0.8076 = 2,229 hours, or 9.3 months at 240 hours a month.

Pick a GPU preset or choose “Custom rate” and type the per-GPU hourly price from your provider. For multi-GPU instances, divide the instance price by its GPU count (AWS’s 8-GPU p5.48xlarge at $55.04 an hour is $6.88 per GPU-hour). Purchase price, power draw and electricity rate only affect the rent-or-buy lines; your electricity rate is on your utility bill.

On-Demand Prices From the Providers, October 22, 2026

RunPod and Lambda prices are from their own pricing pages on October 22, 2026. RunPod’s page has a Community Cloud / Secure Cloud switch; the Secure Cloud price is the one shown by default and is higher for every GPU listed. Pods can be billed per hour or per second. Lambda prices per GPU-hour fall as the instance gets bigger, so an H100 is $4.29 on a 1-GPU instance and $3.99 on an 8-GPU one. The AWS figure comes from the on-demand pricing data behind AWS’s EC2 pricing page (U.S. East, published October 15, 2026): $6.88 an hour for p5.4xlarge with one H100, and $55.04 for p5.48xlarge with eight, which is the same $6.88 per GPU.

Rent or Buy: How the Break-Even Moves

The rental rate dominates. At the Secure Cloud $0.89 an hour, a card used 8 hours a day pays for itself in under a year; run around the clock, in about three months. At the $0.34 Community Cloud price the same card takes more than two years at 8 hours a day. Electricity changes the answer only slightly at U.S. prices.

Budgeting Mistakes and Edge Cases

Paying for idle time. An on-demand instance bills while it runs, whether or not a job is using the GPU. Stopping instances between sessions is often the largest saving.

Ignoring storage. RunPod charges $0.10 per GB a month for volume disks while a pod runs and $0.20 while it is stopped; a 200 GB model cache left idle costs $40 a month.

Comparing whole instances with single GPUs. Hyperscaler prices are often quoted per 8-GPU machine. Divide by the GPU count before comparing.

Counting only the card’s wattage. NVIDIA lists 850 W of required system power for an RTX 4090 build; enter the whole machine’s draw if you want a fairer owning cost.

What the Estimate Does Not Include

It does not include storage, data transfer, the CPU and RAM that come with an instance, or reserved, spot and multi-month pricing, which can be well below on-demand. On the ownership side it ignores the rest of the computer, cooling, resale value, downtime and your time maintaining hardware. It also treats a consumer RTX 4090 and a data-center H100 as interchangeable on price alone; they are not interchangeable for large models that need more memory. For a large purchase or commitment, get a written quote from the provider.

Frequently Asked Questions

How to Use This Calculator

1. Choose a GPU and provider — or select Custom rate and enter the per-GPU hourly price you were quoted.
2. Set GPUs and usage — hours per day and days per month.
3. Add a purchase price — with power draw and your electricity rate.
4. Compare — daily, monthly and yearly cost and the rent-vs-buy break-even.