GPU Rental Profitability Calculator

This GPU rental profitability calculator estimates the monthly profit of a GPU server you rent out by the hour on a compute marketplace, hosted on your premises or in colocation. It works in five steps: Hosting, Rental Income, Rig, Power and Balance, which follows the money month by month over five years and shows the profit at ten prices around the one entered. The results are estimates from the values entered, not a quote, an offer, or financial or tax advice.

GPU rental profitability calculator: from inputs to resultsThe five steps with their inputs and results; the last sets the costs against the net income and finds where the rig breaks even.InputsStepsResultsOn-premises or colocation; networkand rack space ($/month)HostingNetwork, and UPS or rack space, a monthuptime, site costsPrice ($ per GPU-hour), occupancy(%)Rental IncomeRented hours and net income a monthHᵣ, RₙₑₜGPUs, rig price ($), depreciation andinterest rates (%), own capital orfinancingRigDepreciation, interest, insurance andrepairs a monthD, I, U, MPower when rented (kW), electricityprice ($/kWh)PowerEnergy billed and electricity a monthEBalanceProfit a month, break-even price andoccupancy, balance by month

The results can be downloaded as a PNG or copied as text at the end of the page.

Worked Example

The calculator opens with an example: one server with a GeForce RTX 5090, hosted on your premises, listed at $0.80 per GPU-hour and rented 60% of the time. It assumes a rig price of about $7,500, a draw at the wall of 0.9 kW when rented and 0.25 kW when idle, electricity at $0.18 per kWh, an internet line at $40 a month, a UPS at $600, and depreciation of 37% of the rig’s value a year over 3 years. These are typical values, not quotes: replace them with your own quotes, bills and measurements.

Open the calculator

The calculator, with its equations and assumptions, opens from a link we send to your email address. We use the address to send the link, and write to you about our services only if you tick the newsletter box. Details in the privacy policy.

Hosting

This step sets where the rig runs and what the site costs each month.

On your premises the rig runs on your own electricity and internet line, and a UPS (uninterruptible power supply) carries it through short power cuts. In colocation a data centre hosts it: rack space and power are billed each month, and the data centre’s own UPS and cooling are in its prices.

Hosting

On-premises means the rig runs at a site of your own, such as a home, an office or a workshop. Each choice keeps its own values of the inputs that differ between them: uptime, network, setup, electricity price and power overhead, and in colocation the power billing.

On-premises: the internet line, or only what the rig adds to it, such as an upgrade or a second line. Colocation: the network fee, 0 if the rack’s price includes it.

The colocation fee for the rack units the rig takes.

Secondary Inputs

The share of the year the rig is online and listed. Power cuts and line outages lower it.

On-premises: electrical work. Colocation: the installation fee and transport. It is invested with the rig and written off over its life.

The UPS’s price with its batteries.

The price of a replacement set of batteries.

The years the UPS itself lasts.

How often the batteries are replaced.

The share of the power drawn that reaches the rig; the rest is lost as heat in the UPS. The default, 90%, is an assumption: take the figure from your UPS’s data sheet.

Network a month$40
UPS a month$8
Rack space a month$0

Equations

Cnet=12nC_{\text{net}} = 12\,n
Cups=Pups−bNups+bNbC_{\text{ups}} = \frac{P_{\text{ups}} - b}{N_{\text{ups}}} + \frac{b}{N_b}
Crack=12crackC_{\text{rack}} = 12\,c_{\text{rack}}

where CnetC_{\text{net}} is the network’s cost a year and nn its price a month; on your premises CupsC_{\text{ups}} is the UPS’s cost a year: its price PupsP_{\text{ups}} less the batteries bb, over its life NupsN_{\text{ups}} in years, plus the batteries over theirs, NbN_b; in colocation CrackC_{\text{rack}} is the rack space’s cost a year and crackc_{\text{rack}} its price a month; 12 is the number of months in a year.

Network 12 × $40 = $480 a year; UPS (600 − 150) ÷ 8 + 150 ÷ 4 = $94 a year.

The UPS is written off in two parts because they wear out at different rates: the unit over its own life, its batteries over theirs, replaced each time they are spent. APC gives the batteries of its Smart-UPS SRT2200XLA a life of 3 to 5 years (APC, 2026). The network and the rack space are paid every month, the rig rented or not.

Assumptions

  • Nothing is recovered from the UPS or its batteries when they are replaced.
  • The prices a month are the same all year.

Rental Income

This step turns the price and the hours rented into income a month.

Renters pay by the hour for each GPU. The rig earns only in the hours it is rented: the occupancy is the share of the listed hours that are, and the uptime, from Hosting, the share of the year the rig is listed at all. Some marketplaces keep a share of the income as their fee.

The price per GPU per hour the rig is listed at on the marketplace.

The share of the listed hours that renters pay for.

Secondary Inputs

The share of the income the marketplace keeps, with payout costs; 0 if it adds its fee to the renter’s price instead. From the marketplace’s host terms.

What renters pay for storage and data transfer, where the marketplace bills them.

Rented hours a month436 h
Income a month$349
Fee a month$0
Net income a month$349

Equations

Hr=8760arH_r = 8760\,a\,r Rnet=(1−f)(gpHr+12Ro)R_{\text{net}} = (1 - f)\,(g\,p\,H_r + 12\,R_o)

where HrH_r is the number of hours rented a year; 8760 is the number of hours in a year; aa is the uptime and rr the occupancy; RnetR_{\text{net}} is the net income a year: what gg GPUs earn at the price pp in HrH_r hours, plus 12 months of other income RoR_o, less the marketplace fee ff.

Rented hours 8,760 × 99.5% × 60% = 5,230 h a year; income 1 × $0.80 × 5,230 h = $4,184 a year; fee 0%, $0; net income $4,184 a year.

Rgross=gpHr+12RoR_{\text{gross}} = g\,p\,H_r + 12\,R_o Rnet=(1−f)RgrossR_{\text{net}} = (1 - f)\,R_{\text{gross}}

RgrossR_{\text{gross}} is what renters pay in a year; where the marketplace keeps a share ff of it, the net income is that much smaller. The occupancy is the host’s own estimate.

Assumptions

  • Every GPU is rented for the same hours, at the price entered, all year.
  • Reserved-rental discounts and interruptible bids are left out.

Rig

This step spreads the rig’s price over its life and adds what the money and the upkeep cost.

The rig loses value as newer cards arrive: what was invested, less what the rig sells for at the end, is written off over its economic life (depreciation): evenly, or with a yearly rate as a share of its value at the start of each year. The money tied up in it costs interest, charged on the average investment over the life as in the machine hour rate calculator; insurance is charged the same way, and repairs as a share of the investment.

The cards the rig rents out; the price is per card.

The price of the whole machine, cards included, before tax.

Depreciation

Depreciation can be worked out from the economic life or from a yearly rate of the rig’s value; with a rate, the salvage value follows from the rate and the life.

The years the rig is expected to earn before newer cards make it hard to rent.

The share of the rig’s value at the start of each year that is written off in that year (a reducing-balance rate), so the charge is highest in the first year and smaller every year after. Depreciation is charged at this rate over the economic life, and what is left at its end is the salvage value, shown in the results. A straight-line rate, a share of the investment, is not this rate: for it, choose depreciation from the economic life and enter the life it gives, (100% − salvage value) ÷ rate.

Capital

Own capital: the investment is paid at the start. Financing: a share of it is borrowed and repaid in equal monthly payments at the interest rate in Rig; the rest, and on your premises the UPS, are paid at the start.

The cost of the money tied up in the rig: a loan’s rate, or what the money would earn elsewhere. The default is the machine hour rate calculator’s.

Secondary Inputs

The default is the machine hour rate calculator’s.

Fans, power supplies and other parts, and at a colocation paid on-site work, as a share of what was invested.

Salvage value SS$1,89525.00% of the investment
Depreciation a month$158
Interest a month$30
Insurance a month$2
Repairs a month$19

Equations

D=P0−SN,D = \frac{P_0 - S}{N}, S=sP0,S = s\,P_0, P0=P+PsetP_0 = P + P_{\text{set}} AAI=(P0−S)(N+1)2N+S\text{AAI} = \frac{(P_0 - S)(N + 1)}{2N} + S I=iAAI,I = i\,\text{AAI}, U=uAAI,U = u\,\text{AAI}, M=mP0M = m\,P_0

where DD is the depreciation a year: the investment P0P_0, the rig’s price PP with the one-off setup PsetP_{\text{set}}, less its salvage value SS, the share ss of P0P_0, over the economic life NN in years; AAI\text{AAI} is the average annual investment, the rig’s mean value over its life: the investment less its salvage value, times N+1N + 1 over 2 times NN, plus the salvage value; II, UU and MM are the interest, insurance and repairs a year, at the rates ii, uu and mm.

Invested 7,480 + 100 = $7,580, falling each year by 37% of its value for 3 years, to a salvage value of $1,895 (25.00%); depreciation 37% × 7,580 = $2,805 in the first year, smaller every year after, on average (7,580 − 1,895) ÷ 3 = $1,895 a year; average investment $5,121, interest 7% = $358, insurance 0.5% = $26; repairs 3% = $227 a year.

S=P0(1−d)NS = P_0\,(1 - d)^N AAI=P0−SNd\text{AAI} = \frac{P_0 - S}{N\,d}

These are a machine’s fixed costs by the machine rate method, as in the machine hour rate calculator: straight-line depreciation, and interest and insurance on the average annual investment, AAI=(P−S)(N+1)/(2N)+S\text{AAI} = (P - S)(N + 1)/(2N) + S (FAO, 1992, 3.4.2). Fixed costs “do not stop when the work stops” (FAO, 1992, 3.2.1): the rig costs them rented or idle. With a yearly rate, dd is the share of the rig’s value at the start of each year written off in that year, so after nn years the value is P0(1−d)nP_0(1 - d)^n; the salvage value SS is the value left after the economic life NN, DD the yearly depreciation averaged over it, and AAI\text{AAI}, the mean of the values at the start of each year, (P0−S)/(Nd)(P_0 - S)/(N\,d).

Assumptions

  • From the economic life, the rig loses value evenly over it and is worth its salvage value at the end of it.
  • With a yearly rate, depreciation is declining-balance: the rate applies each year to the rig’s value at the start of that year, so the charge is higher in the early years and lower later, and the value left at the end of the economic life is the salvage value (Miyata, 1980, p. 4). The costs a month use the depreciation averaged over the life, and interest and insurance on the average of the values at the start of each year, as from the life. For a life that is not a whole number of years, the same formulas are used with that life. That is this calculator’s convention.
  • Repairs cost the same share of the investment every year.

Power

This step prices the electricity the rig draws, rented or idle.

The rig draws most while it is rented and its cards run at full load, and less while it waits for a renter. Cooling adds to the bill: on your premises as extra air conditioning, in colocation through the data centre’s PUE (power usage effectiveness). On your premises the UPS loses part of the power as heat. In colocation the data centre bills either each kWh the rig draws, or a fixed power reserved for it each month, drawn or not.

Power billing

Metered: each kWh the rig draws. Contracted: the kW reserved, billed for every hour as if drawn, whatever the rig draws. Both at the price per kWh and the PUE.

What the whole rig draws from the wall with its cards at full load: measured with a plug-in meter, or the cards’ rated power plus the rest of the machine.

All in, before tax: energy, delivery and fees. On-premises: your tariff. Colocation: the price per kWh the data centre bills.

The power the data centre reserves for the rig and bills for every hour, whether the rig draws it or not: at least what the rig draws at full load, less any power the rack space’s price already includes.

Secondary Inputs

What the rig draws while listed but not rented, and while offline.

On-premises: 1 plus the share of power spent on extra cooling; 1 without air conditioning. Colocation: the data centre’s PUE, its total power divided by the power its IT equipment draws; 1 where its price includes cooling.

Energy billed a month518 kWh
Electricity a month$93

Equations

E=(WloadHr+WidleHi)keη,E = \frac{(W_{\text{load}}\,H_r + W_{\text{idle}}\,H_i)\,k\,e}{\eta},
Hi=8760−HrH_i = 8760 - H_r
E=8760WckeE = 8760\,W_c\,k\,e

where EE is the electricity’s cost a year; metered, the power when rented WloadW_{\text{load}} for the HrH_r hours rented and the power when idle WidleW_{\text{idle}} for the other HiH_i hours, times the power overhead kk and the electricity price ee, divided by the UPS efficiency η\eta, 1 in colocation; contracted, the contracted power WcW_c for all 8760 hours.

Energy 0.9 kW × 5,230 h + 0.25 kW × 3,530 h = 5,589 kWh a year; × $0.18/kWh × 1 ÷ 90% = $1,118 a year.

NVIDIA gives the GeForce RTX 5090 a total graphics power of 575 W and a required system power of 1,000 W (NVIDIA, 2026): what the card draws at full load, and the power supply it recommends for a whole PC built around it. The power when rented lies between the two: the card at full load, the rest of the machine, and the power supply’s losses.

Assumptions

  • Metered: the rig draws the power when rented in every hour rented and the power when idle in every other hour of the year, offline hours included.
  • The electricity price is the same at every hour of the day.
  • Contracted: the reserved power is billed for every hour of the year, whatever the rig draws.

Balance

This step sets the costs against the net income, finds the price and occupancy at which the rig breaks even, and follows the money month by month over five years.

The profit is the net income less every cost, a year’s worth shown as a month. The cash profit leaves out depreciation and the interest on the money tied up, which are not paid out each month: it is what the rig brings in towards paying for itself, before a loan’s payments, which the balance by month takes off. The break-even price is the price per GPU-hour at which the profit is zero at this occupancy; the break-even occupancy is the occupancy at which it is zero at this price.

Uses the net income from Rental Income and the costs from Hosting, Rig and Power.

Profit a month−$1
Cash profit a month$191
Payback3.564 yearsbeyond its life of 3 years
Cost per rented GPU-hour chc_h$0.8027
Break-even price p*p^{*}$0.8027
Break-even occupancy r*r^{*}60.24%

The money month by month over five years from the purchase, the year’s values spread evenly over its months: the net income and the costs paid to date, the equipment’s value and the debt still owed. The cash balance, in bold, is the money in less the money out; the net position adds the equipment’s value and takes off the debt. The payback is the month the cash balance first covers the debt still owed; it and the end of the economic life are marked where they fall within the five years. The buttons above the chart show or hide each line. Point at the chart, or tap it, to read a month’s values.

Profit a Month by Occupancy and Price

The profit a month at every occupancy from 50% to 100% and every price in the table below, the rest as entered. The point is the occupancy and price entered; the dark line is the break-even, where the rig makes neither a profit nor a loss. Point at the surface, or tap it, to read the profit a month at that occupancy and price. Drag the drawing, or use the arrow keys, to turn it.

Profit a month, $, at each occupancy (rows) and price per GPU-hour (columns)
Occupancy$0.05$0.30$0.55$0.80$1.05$1.30$1.55$1.80$2.05$2.30
50%−322−231−141−5041132223313404495
55%−325−225−125−2674174274374474574
60%−328−219−110−1108217326435544653
65%−331−213−9523141259377495613731
70%−334−207−8047175302429556683810
75%−337−201−6472208344480617753889
80%−340−194−4996241387532677822968
85%−343−188−341202754295847388921,047
90%−345−182−191453084726357999621,125
95%−348−176−31693425146878591,0321,204
100%−351−170121933755577389201,1011,283

Costs a Month

Electricity$9326.63%
Depreciation$15845.14%
Interest$308.540%
Insurance$20.6100%
Repairs$195.417%
Network$4011.43%
UPS$82.233%
Rack space$0

The costs a month by the step they come from, Hosting, Rig and Power, in the inner circle, each split into its parts in the outer ring, clockwise from twelve o’clock. Point at a slice, or tap it, to read its cost and share.

Equations

Π=Rnet−C\Pi = R_{\text{net}} - C
C=E+D+I+U+M+Cnet+CupsC = E + D + I + U + M + C_{\text{net}} + C_{\text{ups}}
C=E+D+I+U+M+Cnet+CrackC = E + D + I + U + M + C_{\text{net}} + C_{\text{rack}}
p*=C/(1−f)−12RogHrp^{*} = \frac{C/(1 - f) - 12\,R_o}{g\,H_r}

where Π\Pi is the profit a year, the net income RnetR_{\text{net}} less the costs CC; CC is the sum of the year’s electricity EE, depreciation DD, interest II, insurance UU, repairs MM and network CnetC_{\text{net}}, with the UPS CupsC_{\text{ups}} on your premises or the rack space CrackC_{\text{rack}} in colocation; p*p^{*} is the break-even price per GPU-hour, at which Π=0\Pi = 0: the costs CC grossed up for the fee ff, less 12 months of other income RoR_o, over gg GPUs and HrH_r hours rented.

Net income $4,184 less costs $4,198 = −$14 a year, −$1 a month. Break-even price (4,198 ÷ 100%) ÷ (1 × 5,230 h) = $0.8027 per GPU-hour.

ch=CgHrc_h = \frac{C}{g\,H_r} r*=F8760aμr^{*} = \frac{F}{8760\,a\,\mu} F=C−E+8760Widlekeη−12(1−f)RoF = C - E + \frac{8760\,W_{\text{idle}}\,k\,e}{\eta} - 12\,(1 - f)\,R_o μ=(1−f)gp−(Wload−Widle)keη\mu = (1 - f)\,g\,p - \frac{(W_{\text{load}} - W_{\text{idle}})\,k\,e}{\eta} tp=P0+PupsΠcasht_p = \frac{P_0 + P_{\text{ups}}}{\Pi_{\text{cash}}}
A=LP0j1−(1+j)−n,A = \frac{L\,P_0\,j}{1 - (1 + j)^{-n}}, j=i12,j = \frac{i}{12}, n=12NLn = 12\,N_L

chc_h is what each rented GPU-hour costs. The break-even occupancy r*r^{*} splits the costs into FF, those that do not change with the hours rented, and μ\mu, the net income each rented hour adds after its energy; with contracted power the energy is all fixed, F=C−12(1−f)RoF = C - 12\,(1 - f)\,R_o and μ=(1−f)gp\mu = (1 - f)\,g\,p. There is no break-even occupancy where μ\mu is not above zero, or where r*r^{*} is above 100%.

The cash profit Πcash\Pi_{\text{cash}} is the profit with the depreciation and the interest added back, and on your premises the UPS’s write-off replaced by its batteries’ replacement; the payback tpt_p is the investment, with the UPS, divided by it, in years. With financing, the payback is when the cash balance first covers the loan still owed, from the balance by month: the loan’s interest delays it. The figure and the table repeat the calculation at each occupancy and price.

The balance by month: the cash balance starts at minus the money put in, the investment and on your premises the UPS, less the loan; each month adds the cash profit a month, Πcash/12\Pi_{\text{cash}}/12, less the loan’s payment AA while it is repaid. The net position adds the equipment’s value, the investment written down to its salvage value over its economic life, evenly or, with a yearly rate, by a share of each year’s value, and on your premises the UPS to its batteries over its own, and takes off the loan still owed. With own capital there is no payment; at a zero interest rate the loan is repaid in equal parts, LP0/nL\,P_0/n.

Assumptions

  • The year’s income and costs are spread evenly over its months.
  • The profit is before income tax.
  • Over the five years the rig and the UPS are bought once and only the UPS batteries are replaced, their cost spread evenly; after its economic life the rig earns and costs as before, its value staying at its salvage value.
  • A loan’s interest rate is the interest rate in Rig, its payments equal and monthly; the profit a month keeps the machine rate’s interest on the average investment.

Cite as: Jacek Garbuliński, GPU Rental Profitability Calculator, Blixild, 2026, https://blixild.com/gpu-rental-profitability-calculator.html

Questions and Answers

When and why should I use this calculator?

Before buying a GPU server to rent out by the hour, or before listing one you already own on a compute marketplace. It shows whether the rent at a given price and occupancy covers what the rig costs to own and run, on your premises or in colocation, and at which price or occupancy it breaks even. It also works the other way round: change the price until the profit a month is what you need, and the table shows the occupancy that price requires.

Is renting out a GPU profitable?

It depends on three things the calculator sets side by side: the price a GPU-hour rents for, the hours it is rented, and what the rig costs to own and run. In the worked example, one RTX 5090 server on your premises about breaks even at $0.80 per GPU-hour and 60% occupancy. Depreciation is its largest cost, close to half the total. A rig bought for less, or earning for longer, breaks even lower.

What occupancy should I enter?

The share of the hours your rig is listed that renters pay for. Use your own listing’s history where you have one; otherwise read the table, which gives the profit at every occupancy from 50% to 100%, at ten prices around yours. The worked example’s 60% is an assumption, not a market figure.

Sources

APC by Schneider Electric (2026). APC Smart-UPS On-Line, 2200VA, Tower, 120V, SRT2200XLA, product data sheet. https://www.apc.com/us/en/product/download-pdf/SRT2200XLA

FAO (1992). Cost Control in Forest Harvesting and Road Construction. FAO Forestry Paper 99. Rome: Food and Agriculture Organization of the United Nations. Chapter 3, Calculation of Machine Rates. https://www.fao.org/4/T0579E/t0579e05.htm

Miyata, E. S. (1980). Determining fixed and operating costs of logging equipment. General Technical Report NC-55. St. Paul, MN: USDA Forest Service, North Central Forest Experiment Station. https://research.fs.usda.gov/treesearch/10120

NVIDIA (2026). GeForce RTX 5090 Graphics Cards, specifications. https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/

Send an enquiry

To rent a GPU machine by the hour instead of buying one, send an enquiry.

Describe your requirements briefly. We respond within one business day.

contact@blixild.com

By submitting this form, you agree to be contacted regarding your enquiry.