---
title: "Why now: the cost of intelligence collapsed | AI Makers"
description: "The cost of intelligence fell 280× in two years. What that changes for your company — and why the window to act is this quarter, not in three years."
url: "https://aimakers.fr/en/why-now"
canonical: "https://aimakers.fr/en/why-now"
language: en
category: "03 Guide / Definition"
cluster: "transformation"
stage: "awareness"
generated: 2026-10-17
source: "Contenu de la page publiée, extrait de son HTML rendu."
---

# Why now.

The cost of intelligence is collapsing. This document explains what that changes for your company, with sourced figures.

01

## Intelligence costs 280 times less than it did two years ago.

In March 2023, GPT-4 cost $30 per million tokens. Sixteen months later, an equivalent model cost $0.15. The Stanford AI Index measures a 280× drop in inference cost over two years. a16z named it LLMflation: at equal performance, the price falls tenfold every year.

÷280

inference cost over 2 years

Stanford AI Index

÷10 a year

at equal performance

a16z

4% → 72%

success on real engineering tasks in one year

SWE-bench

02

## When intelligence is nearly free, your processes become the variable.

Every process that rests on a human reading, deciding and re-keying can be redrawn. At today’s rates, analysing a ten-page document costs about three cents of API. An invoice processed by hand costs $12.88; automated, $2.78 (Ardent Partners).

The question is no longer what AI costs. It’s what each month costs you while your processes stay manual.

$12.88 → $2.78

one invoice, processed by hand then automated (Ardent Partners)

03

## Nine companies in ten use AI. Fewer than four in ten see the effect.

88% of organisations use AI. Only 39% measure an impact on their bottom line (McKinsey, 2025). In France, 55% of small and mid-sized business leaders use generative AI, but only 17% regularly (Bpifrance Le Lab, January 2026). And Gartner predicts more than 40% of agentic-AI projects will be cancelled by the end of 2027: cost, unclear value.

The dividing line isn’t the technology, everyone has the same models. It’s execution. The winners redraw their processes and measure. The losers buy licences and wait.

88% → 39%

of organisations use AI; only 39% measure an impact on their bottom line (McKinsey, 2025)

04

## A lead compounds. So does a lag.

A system in production improves with every piece of data it handles: your playbooks sharpen, your teams level up, your data piles up where your competitors have none. Twelve months of lead can’t be bought back with the same tool — you would have to buy back the twelve months of learning too.

Next year is the most expensive line in your strategic plan.

## This quarter. Not in three years.

You don’t need an 18-month plan to start. Two weeks of audit. At least three costed use cases. First systems in production within 30 days.

[Book my free diagnostic](https://aimakers.fr/en/contact) [See our method](https://aimakers.fr/en#methode)

Sources

-   [a16z, LLMflation (November 2024)](https://a16z.com/llmflation-llm-inference-cost/)
-   [Stanford AI Index 2025](https://hai.stanford.edu/ai-index/2025-ai-index-report)
-   [McKinsey, The State of AI (November 2025)](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai)
-   [Bpifrance Le Lab, 82nd SME business-climate barometer (January 2026)](https://presse.bpifrance.fr/bpifrance-le-lab-presente-le-82eme-barometre-semestriel-de-conjoncture-aupres-des-tpe-pme-les-dirigeants-font-part-dun-leger-regain-de-confiance-pour-2026)
-   [INSEE, ICT survey (2025)](https://www.insee.fr/fr/statistiques/8616837)
-   [Gartner (June 2025)](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027)
-   [Ardent Partners / Medius](https://www.medius.com/resources/guides-reports/ardent-partners-accounts-payable-metrics-that-matter/)
