Skip to main content
NexPatch
Calculator · Cost comparison

Cost of self hosting versus pay per token.

This calculator shows the monthly request volume above which running your own AI infrastructure pays off compared with ongoing pay-per-token billing. It is based on your inputs for request volume, request length, model size, location and term. The result appears immediately, with no form - our own model calculation with disclosed assumptions, not a market study.

Self-hosted
Per request
Break-even
Interactive Cost CalculatorPrice status: September 2026
Req.
20k1 Mio.2,5 Mio.5 Mio.
1,800 tokens
5001.8003.5006.000

Reference price: 1.50 € / Mio. Token

Output Comparison450.0 M tokens / mo
Pay-per-Token (API)
~€675 / mo
Self-Hosted Fixed
~€4,000 / mo
Pay-per-Token (API)€675
Self-Hosted Fixed€4,000
Economic Break-Even Point: ~1,481,000 Req./Mo.

At current volume, API usage is still €3,325 / month cheaper.

Multi-Year Total Cost Projection
API (12 Mo.):
€8,100
Eigenbetrieb:
€48,000
450.0M tokens per monthModel class: medium

Disclosed Assumptions & Hardware Costs

Transparent cost matrix per model class and month (Price status: September 2026):

Model ClassAPI Price / 1M TokensHardware / Month (DC / Cloud)Ops & Support / MonthTotal Self-Hosted
Small (e.g. 8B models)0.30 €1200 € / 1600 €1200 €~2400 € / ~2800 €
Medium (e.g. 70B models)Active1.50 €2200 € / 2900 €1800 €~4000 € / ~4700 €
Large (e.g. 405B / Frontier)4.50 €4500 € / 5800 €2500 €~7000 € / ~8300 €

01 What this calculator answers

Many organizations face the same decision: does it make sense to invest in dedicated, permanent AI infrastructure, or is pay-per-token API billing more cost-effective over time? The answer depends on volume and chosen model size. At low volume, fixed infrastructure costs dominate. At high volume, self-hosting becomes dramatically more cost effective. This calculator shows the exact point where that economic balance tips.

02 What inputs you need

Four inputs give you an initial assessment:

InputMeaning
Monthly request volumeNumber of model requests your use case generates per month
Average request lengthCombined prompt input and output length in tokens
Model sizeModel class (Small 8B, Medium 70B, or Large 405B / Frontier)
Operating environmentOwn data center (On-Premises) or dedicated EU cloud (Frankfurt am Main)

The result appears immediately in the browser, with no forms and no contact details required.

03 How the calculation works

The calculator uses disclosed assumptions (Price status: September 2026) that you can inspect item by item, rather than a black-box formula.

AssumptionBasis
HardwareMarket-standard purchase and depreciation costs for comparable server configurations, our own model calculation
Power and locationAverage commercial electricity price at the selected location, our own model calculation
StaffProportional effort for operating and maintaining your own infrastructure, our own model calculation
Comparison price per tokenPublicly listed prices from common providers, retrieval date shown with each result

Next Steps & Chained Calculators

After clarifying infrastructure costs, calculate process-level financial returns in the ROI calculator or evaluate organisational prerequisites in the AI maturity check.