AI & Automation

Inference Cost Curves and Pricing a Feature That Gets Cheaper Every Year

Set a price for something whose input cost falls steeply and predictably, and understand why cost-plus pricing, which feels prudent, quietly hands the entire cost decline to your customers.

  • Advanced
  • 14 min total
  • 15 chapters

What decision this helps you make: Which pricing model to use for an AI feature, what to promise in a multi-year contract, and whether a falling input cost should show up as a lower price or a wider margin.

What this topic is

An inference cost curve is the path the price of serving one unit of AI work takes over time. It falls, steeply, for two reasons that compound: hardware and serving efficiency improve, and the models needed for a given task get smaller as capability at every size improves. The pricing question follows: if the cost of delivering your feature falls by half or more each year, what should happen to the price? The answer depends entirely on which comparison your customer is making, and that is a choice you make rather than a fact you discover.

Why it matters

This is the rare situation where a business gets a large, predictable annual cost reduction, and most companies give it away without noticing. Prices are sticky upward and slippery downward, so a decision made in year one persists. A price anchored to cost falls with cost forever, while a price anchored to the value of the job done keeps the decline as margin. The same cost curve therefore produces either a business with steadily widening gross margins or one whose revenue per unit shrinks faster than volume grows.

Who should learn it

Founders and product leaders pricing an AI feature, operators renegotiating a vendor contract, and anyone signing a multi-year commitment on either side of an inference bill.

What you will understand

  • Why costs fall on a learning curve rather than a calendar, and what that implies for forecasting
  • How the same cost decline produces expanding margin or collapsing revenue, depending on the pricing model
  • Why cheaper inference usually raises your total bill rather than lowering it
  • What to write into a contract on each side when both parties know the input price is falling

Prerequisites

Common misconception

"Costs are falling, so we should pass the savings on and win on price." Passing savings on is a strategy, not a default, and the default version of it is the expensive one: cost-plus pricing hands the customer every future decline automatically, permanently, and without you getting anything for it. Deliberate price cuts are different: they buy volume, market share or a competitor's customers, and they are decided each time with a number attached. The distinction matters because a price that falls automatically also teaches your customers to wait, which is the one behaviour a subscription business cannot afford to train.