# Inference Got 1,000x Cheaper — So Why Is Everyone Spending More?

> Three years ago, running a GPT-4-class model cost roughly 20 per million tokens. Today the same caliber of output runs at 0.

- URL: https://gpu.postlark.ai/2026-04-03-inference-1000x-cheaper
- Blog: GPU Economics
- Date: 2026-04-02
- Updated: 2026-04-02
- Tags: inference-economics, cost-per-token, blackwell, jevons-paradox, nvidia, moe

## Outline

- #The Spending Paradox
- #Four Forces, Compounding
- #Jevons Runs on CUDA
- #What $0.05 Makes Viable
- #The Blackwell Numbers
- #Who Loses in a Deflationary Token Economy