GPU hosting profitability calculator

Is renting out your GPU worth it?

GPU rental marketplaces let owners of idle cards earn per-hour payouts from people training and running AI models. Whether that pays depends on four numbers: how much the platform pays per GPU-hour after its fee, what share of the hours you actually get rented, how much power the card draws, and what your electricity costs. This calculator puts them together and tells you net profit per hour and per month, how long until the hardware pays for itself, and the utilization you need just to cover power.

Start by picking a GPU preset to fill in board and idle power, then set your electricity price from your bill (total charges divided by kWh, including delivery) and a payout rate from current listings on your platform. Everything is editable — presets are estimates, not measurements.

GPU presets

The formula

power $/h (rented) = (N · P_board + P_system) / 1000 · price · PUE
power $/h (idle)   = (N · P_idle  + P_system) / 1000 · price · PUE
avg power $/h      = u · rented + (1 − u) · idle
revenue $/h        = u · N · payout · (1 − fee)
net $/h            = revenue − avg power
net $/month        = net $/h · 730
break-even months  = hardware cost / net $/month

N is the number of GPUs, u the utilization (0–1), and 730 the average hours in a month (8,760 ÷ 12).

Worked example

One RTX 3090 (350 W) in a system drawing another 80 W, electricity at $0.15/kWh, PUE 1.1, rented 60 % of the time at $0.25 per GPU-hour with a 20 % platform fee, bought for $900. The card alone at full load costs 350 ÷ 1000 × 0.15 = $0.0525 per hour before overhead. Including the system and PUE, a rented hour costs $0.0710 and an idle hour $0.0182. Average revenue is 0.6 × 0.25 × 0.8 = $0.1200 per hour, so net is $0.0702 per hour or $51.22 per month, paying back the card in about 17.6 months.

Frequently asked questions

How much does it cost to run a GPU 24/7?

Power cost per hour is watts ÷ 1,000 × your price per kWh. A 350 W card at $0.15/kWh costs $0.0525 per hour, or about $38.32 per month flat out — before cooling overhead and the rest of the system.

What utilization should I assume?

Rental demand is uneven. New hosts with no reliability history often see low utilization at first. Try 30 %, 60 % and 90 % to see how sensitive your result is — the break-even utilization line tells you the minimum you need just to cover power.

What is PUE and what should I enter?

Power usage effectiveness is total facility power divided by IT power. A rig in an open basement is close to 1.0; if you run extra air conditioning to remove the heat, 1.1–1.3 is realistic. Commercial data centers average about 1.5–1.6.

Are the GPU presets accurate?

They are manufacturer board-power ratings (TGP/TDP) and rough idle figures — editable estimates. Real draw depends on the workload, power limits and clocks. A plug-in watt meter gives you the true number.

Does this include taxes, internet or hardware wear?

No. Rental income is usually taxable, your internet plan may have usage limits, and cards run hard 24/7 can fail sooner. Keep a margin above break-even for these.

Sources

1. NVIDIA, AMD and Intel product specification pages (total graphics power / board power, TDP / default TDP, core counts).
2. U.S. Energy Information Administration. Electric Power Monthly, Table 5.6.A — Average Price of Electricity to Ultimate Customers by End-Use Sector. www.eia.gov
3. Uptime Institute. Global Data Center Survey (annual). Industry-average power usage effectiveness (PUE). uptimeinstitute.com
4. The Green Grid (2012). WP#49 — PUE: A Comprehensive Examination of the Metric.