Calculators › GPU TCO & $/GPU-hour

GPU TCO & $/GPU-hour

Estimate owned cost per productive GPU-hour — server capex over its depreciation life, installed-to-active allocation, electricity at TDP × PUE, opex, and a facility charge per kW of IT — then compare against the rate you would actually sign: a contract or reserved rental (H100 ~$2.00/GPU-hr, the SemiAnalysis InferenceX 'retail' tier), not the on-demand list. Linked to your project, the facility charge is your own building's capital recovery (about $147/kW-month for the default GB300 rack); on its own the calculator starts at $210/kW-month, inside CBRE's Q1 2026 Northern Virginia asking range of $190–235.

ScenarioNVIDIA GB300 NVL72 · 56 active / 72 installed GPUs · 1 racks · 0.14 MW IT · $9M programguide defaults (as of 2026-09)Edit your project assumptions →

GPU TCO & cost-per-GPU-hour

Server cost$From your project: $86,800 per GPU × 72, a planning value from one March 2026 report of $6.0–6.5M/rack that cites anonymous sources; replace it with your quote. Complete system price for the GPUs in the next field. Use your quote.GPUs / serverGPUsDepreciation lifeyr5 years is the middle of the 4–6 year GPU economic life the guide uses (Goldman Sachs); 2–3 years is the contested bear case — test 3/5/7GPU TDPWServer power multiplier× GPU TDPFrom your project: 135 kW rack ÷ (72 GPUs × 1,400 W). Total server IT power ÷ total GPU TDP: 30% host overhead is 1.30, not 0.30.Annual-average IT load% of nominalIndependent of productive utilization; use the electrical load study.PUEratio≥1 by definitionPower price$/kWhOwned-use utilization%Annual productive hours used to recover ownership costInstalled / active allocation×Installed GPUs ÷ active GPUs; allocates stranded rack slots to productive unitsOpex%/yrFacility / colocation charge$/kW-monthFrom your project: your building's own cost, $1.59M facility over 15 years plus $0.66M network over 5 years, per kW of IT. A lease rate belongs here only if you rent the space. Default $210: inside CBRE's Northern Virginia Q1 2026 asking range of $190–235 for 250–500 kW, energy billed separately. For a hall you build, enter its capital recovery per kW-month instead; 0 only if the building is already paid off.Contract rental comparator$/GPU-hrThe rate you would actually sign. SemiAnalysis InferenceX 'retail' tier, Sep 2026: H100 ~$2.00, GB200 ~$4, GB300 ~$5; Silicon Data's neocloud H100 index read ~$2.8/hr on 2026-10-05 (one dated reading). On-demand list runs far higher (H100 ~$3.5, GB200 ~$16, GB300 ~$18): context, not a comparator.

$6.11 / active GPU-hr · server ownership, energy and facility charge

Annual cost / active GPU$37,450

· Depreciation$22,320

· Energy$1,943

· Opex$8,928

· Facility charge$4,259

· Installed-to-active allocation1.286× across all four cost lines

Breakeven utilization vs contract rental86%

The facility charge puts space and power delivery into the owned cost, so it compares like-for-like with a rental rate that already includes them; energy is billed separately at PUE. Compare against the contract rate you would actually sign, not an on-demand list price. Ownership beats that rate only above the breakeven utilization computed from your own inputs — below it a debt-financed cluster bleeds cash → Ch 1.8.

Worked decision — stated assumptions

All ledger inputs are assumed teaching values chosen to exercise this chapter’s method; they are not market quotes, OEM performance claims or operating records. Arithmetic uses the full entered precision.

Assumed inputs
Input	Value
Server cost	240000 $
GPUs / server	8 GPUs
Depreciation life	3 yr
GPU TDP	1000 W
Server power multiplier	1 × GPU TDP
PUE	1 ratio
Annual-average IT load	100 % of nominal
Power price	0.1142 $/kWh
Owned-use utilization	50 %
Installed / active allocation	1.25 ×
Opex	5 %/yr
Facility / colocation charge	0 $/kW-month
Contract rental comparator	4 $/GPU-hr

Illustrative server-ownership decision: assume $30,000 per installed GPU, three-year straight-line cost allocation, $1,000 annual energy, annual opex equal to 5% of purchase cost, and 1.25 installed GPUs per active GPU. Annual cost per active GPU is (30,000 / 3 + 1,000 + 1,500) × 1.25 = $15,625. At 50% productive use, divide by 4,380 hours: $3.57 per productive GPU-hour.

An assumed like-for-like $4/hour rental crosses that cost at 15,625 / (8,760 × 4) = 44.6% utilization. Above that utilization ownership is cheaper on these costs; below it rental is. With all other inputs fixed, five- and seven-year ownership lives reduce the result to $2.43 and $1.94/hour. These are life sensitivities, not evidence that the fleet earns for those periods. The facility charge is set to zero here to isolate the server; a rental rate includes powered, cooled space, so enter the facility's capital recovery per kW-month and reconcile services outside either quote before making a full-service purchase decision.

At 40% productive use, ownership costs $4.46 per productive GPU-hour and the assumed $4 rental wins. Change productive use, not the independent electrical-load assumption.

Guide derivation. Method and handoff: Chapter 1.8 — Business Models, Economics & ROI.

These are transparent screening estimates, not final designs, financial advice, or project approvals. Replace every default with current project data and have the responsible project authorities validate the result. You can save, share a permalink, or export to CSV; see the full calculator suite.