Skip to content
← Back to blog
AI Strategy·June 22, 2026·6 min read

The AI cost curve: cheap to start, expensive to keep

Inference got 280× cheaper in 18 months — so why do most AI projects lose money? The cost isn’t in starting. It’s in everything after.

There's a seductive promise in AI-first development: ship in an afternoon what used to take a quarter. The data says it's half-true — and the missing half is where budgets die.

Starting has never been cheaper

Per Stanford's 2025 AI Index, GPT-3.5-level inference fell from $20 to $0.07 per million tokens between late 2022 and late 2024 — a 280-fold drop in 18 months. Prototyping is genuinely cheap.

Keeping it is where the bill arrives

Traditional development front-loads the pain: high upfront cost, low tail. AI-first development inverts it — cheap to start, and the meter never stops.

The pattern that works

  1. 1.AI at the edges, deterministic code at the core.
  2. 2.Build the eval harness before you scale.
  3. 3.Abstract the vendor so swapping models is config, not a rewrite.
  4. 4.Budget the tail, not just the launch.

Sources

Written by ivector
Start a project →