---
id: "concept-scaling-laws-valuation"
type: "concept"
source_timestamps: ["¶5", "§ Ask the Bot"]
tags: ["scaling-laws", "ai-economics", "compute-vs-data"]
related: ["claim-data-value-percentage", "framework-cmo-compensation", "concept-data-mixture-weights", "prereq-scaling-laws", "entity-metr"]
definition: "Empirical regularities predicting model performance based on compute and data volume, which can be used to calculate the total percentage of a model's value attributable to training data."
sources: ["tail1"]
sourceVaultSlug: "hbr-seg-tail1"
originDay: 1
articleStem: "hbr-tail-109-ai-pay-fair-rates-content"
sourceUrl: "https://hbr.org/2026/06/how-ai-companies-can-pay-fair-rates-for-the-content-they-need"
sourceTitle: "How AI Companies Can Pay Fair Rates for the Content They Need"
---
# Scaling Laws for Aggregate Data Valuation

## Definition

Scaling laws are the empirical, highly regular relationships between an AI model's performance and its two primary inputs: **compute** (model size) and **training data volume**. From an economic perspective, a scaling law functions as a **production function** mapping inputs to outputs.

## Why it matters for compensation

Scaling laws let researchers **isolate** what share of a model's total pre-training value is attributable to training data versus computational power and algorithmic innovation. This is the basis for [[claim-data-value-percentage|the 20–50% data-value estimate]] and drives **Step 1** of the [[framework-cmo-compensation]] (set the total payment pool).

Two properties make the metric attractive:
1. It relies on **existing evidence** rather than requiring new experiments.
2. It **automatically updates** as technology shifts — e.g., if models become more data-hungry, the data share rises.

Pairs with [[concept-data-mixture-weights]]: scaling laws size the total pool, mixture weights split it. Independent bodies like [[entity-metr]] are proposed to estimate the scaling-law-implied share so firms cannot understate it.

## Prerequisites & caveats

Understanding this requires [[prereq-scaling-laws]] (e.g., familiarity with Chinchilla-style laws). **Enrichment caveat:** while the technical literature confirms scaling laws exist and are useful for *optimization*, it does not by itself prove they are sufficient to *set market prices* or resolve distribution among individual creators; the specific numeric bounds are treated as unverified by the sources reviewed.