Self-Hosting an LLM vs Using an API: Breakeven Explained

Self-hosting looks cheap when GPUs are busy and expensive when they sit idle. The breakeven depends on utilization more than anything else.

Quick answer

Self-hosting wins only at high, steady utilization on a model that meets your quality bar, after counting engineering time.

The inputs

GPU cost per hour.
Throughput in tokens per second for your model and setup.
Average utilization across the day.
Engineering time to run, update and scale it.

The formula

Cost per 1M tokens = GPU cost per hour ÷ (tokens per second × 3,600 × utilization) × 1,000,000. Compare that with the API's blended price per 1M tokens.

Why utilization dominates

At 20% utilization you pay for five times the capacity you use. APIs bill only what you use, which is why they usually win for spiky traffic.

Before you self-host

Make sure caching, batching, output limits and cheaper hosted models haven't already closed the gap. Then run the numbers in the self-host vs API calculator.

Frequently asked questions

Is self-hosting an LLM cheaper than an API?

Only at high, steady utilization with a model that meets your quality bar, after engineering time.

How do I calculate LLM self-hosting cost per token?

Divide GPU cost per hour by tokens produced per hour at your real utilization.

Want us to find your 20%?

Book a free 15-minute API cost teardown. Bring last month's provider usage and we'll show you your biggest waste and the first fix we'd ship.

Get a Free 15-Minute API Cost Teardown