Local AI vs API

Amortized hardware plus electricity versus API pay-as-you-go, so you can see the volume where self-hosting starts to pay.

GPU / hardware cost

USD

Amortize over

months

Power draw

watts

Electricity price

$/kWh

Active hours per day

Monthly volume

M tokens

API model to compare against

$188

Self-hosted: $167 hardware amortization + $21.60 electricity per month. Your time is not included.

$6.88

50M tokens/month on Qwen Flash (75% input / 25% output).

Self-hosting wins on price only when the hardware stays busy. Add your own value for setup and maintenance time before deciding.

Uses rates verified 2026-08-31. Estimates only — actual bills depend on tokenizer behavior, caching hit rates, and provider-side rounding.