---
id: "claim-data-valuation-feasible"
type: "claim"
source_timestamps: ["¶3", "¶4", "§ A Sustainable Future"]
tags: ["data-valuation", "technical-feasibility"]
related: ["concept-data-mixture-weights", "concept-scaling-laws-valuation", "concept-equimarginal-principle", "contrarian-data-valuation-possible", "entity-chris-olah", "entity-dario-amodei"]
confidence: "high"
testable: true
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"
---
# Valuing AI training data at scale is technically feasible and low-cost

## Claim

Valuing AI training data at scale is **technically feasible and low-cost** — directly refuting the industry defense that compensating millions of creators is impossible because the cost of valuing individual data would swallow the value created (see [[quote-data-valuation-objection]]).

## Support

- Low-cost methods have existed **since at least 2021**, evidenced by internal Anthropic documents attributed to [[entity-chris-olah|Chris Olah]] and [[entity-dario-amodei|Dario Amodei]] surfaced in legal discovery.
- Because builders already calculate [[concept-data-mixture-weights]] and [[concept-scaling-laws-valuation|scaling laws]] as a mandatory part of training, the metrics for both **relative** and **aggregate** data value are produced "as a matter of course" and cost nothing extra.
- The economic bridge is the [[concept-equimarginal-principle]].

This is the flagship of the contrarian thesis [[contrarian-data-valuation-possible]].

## Confidence: HIGH · Testable: yes

## Enrichment caveat

Partially validated. The literature (Apple's *Scaling laws for optimal data mixtures*) confirms mixture weights and scaling laws exist and support optimization. But the stronger step — that these signals make valuation "without extra cost" and sufficient to set prices, audit, and distribute among individuals — is a **policy inference**, not an established fact.