Why the gpu inference cost calculator matters
Self-hosting is a fixed-cost bet against a variable-cost API. It wins only when utilisation stays high, which is exactly what a peak buffer works against. This page turns that decision into a handful of inputs you can defend in a budget review: volume, unit cost, rate of adoption, and time. The output is a planning baseline, not a promise — it tells you whether the idea deserves a vendor quote, a pilot, or a pass.
- • Biggest swing factor: sustained utilisation
- • Second-order factor: throughput per GPU, which depends on model size and batching
- • Often ignored: the peak buffer, which is pure idle cost
What actually changes the answer
sustained utilisation moves this number first, then throughput per GPU, which depends on model size and batching. Run a conservative case and an upside case before you commit. If the maths only works in the upside case, treat it as a time-boxed test with a kill date rather than a line in next year's plan.
What to do with the result
Compare cost per request here against your API provider's per-request price. Below 50% utilisation, stay on the API and revisit when traffic is steadier.
Related guides
Long-form playbooks on the same topic, written by the RevenueLab editorial team.
LLM Token Costs in 2026: Pricing Every Model, Hidden Multipliers, and Margin Math
Input vs output token pricing across GPT, Claude, and Gemini, the context-window cost trap, how caching and batching cut bills 40–80%, and the real per-user margin most AI apps miss.
Read the guideSales Capacity Planning: Targets, Quotas, Ramp and Hiring
A monthly cohort model for sales capacity that accounts for expected attainment, ramp curves, and attrition — with a 12-month worked example and the four sanity checks that catch bad plans.
Read the guideFAQ
What does the gpu inference cost calculator work out?
It applies Units = ceil(demand × (1 + buffer) ÷ throughput per unit) to the values you enter for requests per month, requests per gpu per month, monthly cost per gpu, peak headroom buffer. Self-hosting is a fixed-cost bet against a variable-cost API. It wins only when utilisation stays high, which is exactly what a peak buffer works against.
How accurate is this gpu inference cost calculator?
Capacity maths is exact; throughput per GPU is the estimate. Benchmark your own model and batch size before committing to reserved hardware. Replace the defaults with your own invoice, usage export, payroll data, statement, or vendor quote before making a commitment — the maths is exact, so the answer is only as good as the inputs you feed it.
Which input should I stress-test first?
sustained utilisation. Re-run with a pessimistic value for it; if the decision flips, that assumption is the thing you need real data on before signing anything. After that, check throughput per GPU, which depends on model size and batching and the peak buffer, which is pure idle cost.
Which scenario should I start from?
Start with the preset closest to your situation — lean case, expected case, scaled case — then edit the sliders. Presets are realistic starting points, not benchmarks to match, and every change updates the result instantly.
What should I do after running the numbers?
Compare cost per request here against your API provider's per-request price. Below 50% utilisation, stay on the API and revisit when traffic is steadier. A useful planning benchmark to compare against: Self-hosting beats API pricing above roughly 60–70% sustained GPU utilisation.
Can I share or save this calculation?
Yes. Your inputs are written into the page URL, so copying the link shares the exact scenario you are looking at — the person who opens it sees the same numbers. You can also export the inputs and results to CSV or PDF from the result card and keep it with the rest of your workings.
How this calculator is built
Independently maintained
Written by Sam Doshi and the RevenueLab editorial team. We don't sell the data feeds this tool is built on.
Sourced from primary data
Benchmarks come from public AdSense / Stripe / IRS disclosures and reader-submitted data — never third-party "$X per view" claims. Full methodology.
Last editorial review
Reviewed on a rolling quarterly cycle. Dated reviews are published on the methodology record for each calculator.
Editorial standards
See our editorial policy and disclaimer. Results are estimates, not advice.