Skip to content
Sign in
Rung 13 Inference & Token-Pricing EconomicsSwitch rung
Rung 13 / What a task actually costs

Inference & Token-Pricing Economics

The economics of serving models — what a token costs to make, and what it sells for.

17
Sources
8
Concepts
5
Entities
A small precision balance with a copper weight on one pan, on paper.

In scope: token pricing trends, inference gross margins, the AI-lab P&L question, and how memory, optics and foundry cost structure feed through to the price of a token.

A brass-and-paper gauge with a copper bezel: token price in, task price out.
Token pricesTCOMarginLock-inRugs
Download Report

Cut the Cost Per Task

Allocate engineering effort across the four measured 2026 inference-efficiency levers — KV-cache, quantization, speculative decoding, batching

Priority MatrixIntermediate4m

Three Neoclouds

Financial forensics on three fictional GPU-rental providers — one is selling compute below true cost. Find it and flag the evidence

The AuditIntermediate6m