GPU Rental Income: Build a Budget From Paid Hours

On this page 7 sections

A GPU listing is an offer to rent hardware. It is not evidence that the hardware will be rented, or that the listed hourly price will reach the host’s bank account. Before buying a card for hosting, work out how many paid hours would be needed to cover electricity, fees and upkeep. Then ask what evidence supports that amount of demand for your particular machine.

This guide provides a reproducible budget rather than a monthly earnings promise. We have not operated the hosts described below, surveyed host utilization or established a representative market payout. All financial inputs in the worked comparison are explicitly chosen assumptions. Change them in the GPU rental income calculator to test your own offer.

What a host actually commits to

Vast.ai’s hosting overview describes responsibilities that extend beyond installing a graphics card: machine setup, troubleshooting, managing pricing and scheduling maintenance around rental contracts. It explains that an accepted rental locks in its terms and that unlisting an offer does not end an existing contract. Check the current hosting agreement before making a commitment.

Those requirements matter to the budget. A computer you want to use for gaming on demand may not be available to fulfill a renter’s contract. Network access, storage, CPU memory, cooling and uptime affect what can be offered. A comparison between GPU models that ignores the rest of the machine does not establish which host will earn more.

Server rack with blue lighting in a data center room
Stock photograph illustrating server equipment; it does not document a host tested by Cost of Owning.

Define the hourly rate before multiplying

There are at least three different numbers that can be called a rate: a public asking price, a contracted price for booked time, and the host receipt after platform deductions. Use one basis consistently.

If an hourly rate is before fees, subtract the platform’s applicable fee once. If it is already the after-fee host receipt, do not subtract that fee again. A generic percentage reduction from a public listing is not a substitute for reading your agreement or reconciling settled transactions. Our calculator offers both bases and disables its extra fee when you choose after-fee receipts.

Calculate paid utilization as paid GPU-hours divided by the hours for which that GPU is included in your reporting period. For one GPU in a 30-day month, the denominator is 720 hours. A four-GPU machine needs a consistent denominator of 2,880 GPU-hours for that month. Keep availability and payment records aligned; online time alone is not paid utilization.

A worked monthly budget

Assume one GPU, a 30-day month and the following inputs. None is a claimed market average or a recommendation for a particular card:

  • Hourly rate before fees: $0.20.
  • Platform fee: 10%, a scenario assumption.
  • Paid utilization: 30%, a scenario assumption.
  • Whole-system idle power: 80 W; rented power: 350 W.
  • Electricity price: $0.20/kWh.
  • Other monthly costs: $5.
  • Hardware and supporting equipment purchased for hosting: $1,200.

Paid time is 720 × 0.30 = 216 hours. Rental revenue is 216 × $0.20 = $43.20; fees remove $4.32, leaving $38.88 in receipts.

The machine spends 216 hours at 350 W and 504 hours at 80 W. Energy is (216 × 350 + 504 × 80) ÷ 1,000 = 115.92 kWh. Electricity costs $23.18. Subtract that and the $5 allowance from receipts: the operating surplus is $10.70.

This surplus excludes hardware depreciation, financing, tax and unpaid work. Dividing $1,200 by the unrounded monthly surplus gives approximately 112 months of simple payback. That is a constant-input calculation, not a prediction that the machine will remain rentable for that long.

Change one assumption at a time

Hold the hourly rate, fee, power states, electricity price and $5 allowance constant. Changing paid utilization produces these calculated outcomes:

Paid utilizationPaid hoursRevenue before feesElectricityOperating surplus
0%0$0.00$11.52−$16.52
15%108$21.60$17.35−$2.91
30%216$43.20$23.18$10.70
50%360$72.00$30.96$28.84
75%540$108.00$40.68$51.52

These are five scenarios for the same hypothetical machine, not five observations. There is no basis here for calling 30% “typical” or treating higher utilization as attainable. The zero-booking case makes a particular risk visible: keeping a host available still costs money when nobody pays to use it.

Solve for the occupancy that covers running costs

For each rented hour in this example, receipts after fees are $0.18. Extra power above idle costs (350 − 80) ÷ 1,000 × $0.20 = $0.054. The contribution toward standing costs is therefore $0.126 per paid hour.

Idle electricity plus the other-cost allowance is $11.52 + $5 = $16.52. Covering it requires $16.52 ÷ $0.126 = 131.11 paid hours, or approximately 18.2% utilization. This threshold covers operating costs only. It does not repay the hardware.

If the required share exceeds 100%, that setup cannot recover its running costs under those assumptions. If after-fee hourly receipts do not exceed the extra rented-hour electricity cost, additional bookings do not provide a positive contribution toward standing expenses. The calculator flags those cases instead of displaying an attractive payback number.

Home desktop PC setup with dual monitors on a desk
Stock photograph; the example’s power readings are assumptions, not measurements of this computer.

Replace assumptions with your own records

Keep a simple monthly reconciliation: paid GPU-hours, settled rental receipts, platform deductions, storage and bandwidth receipts or charges, wall-meter kWh, additional expenses and time spent maintaining the host. Compare the payout period with the energy measurement period, including any payout delay.

Use whole-system power, not the GPU’s advertised maximum rating. Measure both idle availability and a representative rented workload where possible. A meter reading can incorporate CPU, memory, fans and power-supply losses. Cooling outside the computer may need a separate allowance. The Energy Department’s appliance energy guidance explains the watts-times-hours approach and the usefulness of electricity monitors.

For a machine already running for another purpose, compare incremental hosting costs with its existing baseline. For a machine kept on specifically to attract renters, include its full available-time electricity. State which comparison you are making; calling electricity a cost you “pay anyway” can conceal the extra hours that hosting requires.

FAQ

How much will my GPU earn each month?

We cannot infer booked demand or receipts from the model name. Enter your contracted or settled hourly receipt and paid utilization, then compare several lower-demand cases. The worked example is arithmetic, not a claim about expected earnings.

Can positive operating income still be a loss?

Yes. The $10.70 example leaves out depreciation, financing, tax and labor. A hosting investment can lose value even while its monthly receipts exceed electricity and other operating costs.

Is hosting always better than mining?

No. Use comparable periods, measured power, realized receipts and all applicable fees. Mining adds coin and network conditions; hosting adds booking and contract risks. The GPU mining calculator models a separate scenario, not an interchangeable guaranteed income stream.

What changed in this guide?

October 25, 2026: removed unsupported “realistic” utilization, earnings-by-card and purported host-outcome claims. Replaced them with a labeled budget, a utilization sensitivity comparison and an operating break-even calculation. The calculator now distinguishes before- and after-fee rates and separates idle from rented power.

Published by Cost of Owning. Sources, chosen inputs and worked examples serve different purposes. Check the links and assumptions next to the claim before using a number in your budget.

Suggest a correction with this page’s URL, the figure and a supporting source. Read our editorial and correction policy.