H

How Much To Run AI

Self-hosted open-weight LLM cost calculator comparing owned hardware, rented GPU, and cloud API running costs

free
Last Updated 10/27/2026

About How Much To Run AI

Self-hosting cost calculator for open-weight LLMs: pick models such as Kimi, GLM, DeepSeek, and Qwen to compare the cost of running them on your own hardware or on rented GPUs. Enter the machine's power draw, runtime, and price per kWh to get an electricity cost estimate, which replaces the in-house hardware total cost estimate; hosted API costs are estimated from stored rates and total monthly tokens, with peak and off-peak rates forming a range. The model catalog lists 72 models, of which 2–3 can be compared side by side, using fields from dated catalog records. The credit pack page lists three tiers — Starter $10, Pro $100, and Volume $1,000 — currently marked as not purchasable.

Best for

AI infrastructure teams running self-built GPU clusters
Platform engineers responsible for private LLM deployments
Technical decision-makers evaluating self-hosting versus buying API access
IT cost control staff focused on compute spending

Best use cases

Estimate the monthly electricity cost of running open-source models on owned servers

Compare the break-even point between buying GPU hardware and renting cloud GPUs

Estimate the monthly cost range of hosted APIs billed per token

Compare several candidate models side by side during selection

Features

Whole-machine electricity cost estimate

Calculates the electricity cost of the whole machine from the entered power, runtime, and price per kWh; this result replaces the in-house hardware total cost estimate and must not be added to it.

Owned versus rented comparison

Shows purchased hardware alongside rented GPU and hosted API costs; when a public cloud price matches, it gives the break-even point against renting.

Model catalog and side-by-side comparison

The model catalog lists 72 models, of which 2–3 can be compared, using fields from dated catalog records; the billing units of different APIs are shown separately.

Credit pack tiers at a glance

Starter $10 (100,000 credits), Pro $100 (1,000,000 credits), Volume $1,000 (10,000,000 credits); credits do not expire, and the page marks them as not purchasable.

Pros

  • Simple inputs: power, duration, and electricity price give the whole-machine electricity cost
  • Covers three cost bases at once: owned hardware, rented GPU, and hosted API
  • The model catalog holds 72 models, with 2–3 selectable for side-by-side comparison
  • The cost calculator is free and open to use

Cons

  • The credit pack page marks the Starter, Pro, and Volume tiers as currently not purchasable
  • Hardware with no matching public cloud price cannot produce a break-even point against renting

FAQ

It is calculated from the whole machine's electricity use over the entered number of hours and covers electricity only; equipment purchase, depreciation, labor, and cloud rental are excluded. This result replaces the in-house hardware total cost estimate and must not be added to it.
When a hardware entry has no matching public cloud price, the page cannot estimate a break-even point against renting; the comparison appears once a public cloud price matches.

Tool Information

DeveloperHow Much To Run AI
Release Date10/27/2026
Pricingfree