Dataset·Ledger

AI Spend per Employee

Name: AI Spend per Employee
Creator: Hermosa AI
Published: 2026-10-24
License: https://creativecommons.org/licenses/by/4.0/

Published AI spend figures differ by three orders of magnitude, and most of the disagreement is not disagreement at all: it is six different quantities wearing one label. This ledger holds one row per published figure, with every axis of its own denominator, and converts to a common monthly per-employee basis only where that conversion is defensible. Browse it below, or take the raw files.

Snapshot: 2026-10-24
Coverage: 8figures·3publishers·7on the common axis·5comparability groups
Download

The spread

Every figure that can share one axis. US dollars per employee per month, log scale. Annual figures are divided by twelve, which is safe; nothing else is converted. Points are coloured by what the number is, not by who published it.

Not on the chart, and why

Anthropic: Claude Code enterprise average per developer $150 to $250 per developer per monthNot harmonisable. Converting per developer to per employee needs a developer-to-headcount ratio, which neither Anthropic nor the adopting organisations publish. The value field stores the midpoint of the published band purely for sorting; the band is what was published.

Sharing an axis is not the same as being comparable. The two Atlanta Fed points are an employment-weighted mean that includes firms spending nothing; the Ramp points are unweighted medians among firms that already pay for AI. They sit on one scale here so the spread is visible, and the table below records every axis on which they differ.

About this dataset

[show]

Unit: One row = one published figure, never one source. A publisher that states several cohorts, vintages or years yields several rows: the Atlanta Fed alone yields about eight.
The denominator: Recorded on every row across seven axes: what the dollar is divided by, the population, whether firms spending nothing are inside the denominator, the weighting, the statistic, the scope of the numerator, and the basis. Two figures are comparable only when all of them match, which is what the comparability key records.
Harmonisation: Annual to monthly is safe division and is applied. Per seat or per developer to per employee needs a penetration or staffing ratio that is almost never published, and a mean cannot be turned into a median at all. Those rows keep an empty harmonised value and state the reason, rather than manufacturing a false reconciliation.
Why the famous range is misleading: The widely quoted span of about twelve dollars to several thousand dollars per employee per month is not a dispute between sources. Both ends come from one Ramp release, one month, one definition: the median adopting firm and the median of the top one percent cohort. They reconcile perfectly, because they describe one very skewed distribution.
The collision to watch: The Atlanta Fed reports a top decile at $7,480 per employee per year. Ramp reports a top centile at $7,400 per employee per month. Nearly identical digits, twelve times apart in period, different cohort, different statistic, different basis.
Sources: Every seed figure was read in its primary document: the NBER working paper PDF, both Ramp vintages, and the vendor documentation. Where a figure is reachable only through a summary, the row says so.
License: Free to use, distribute, and reproduce with attribution to Hermosa Labs LLC (CC BY 4.0).
Cite: Hermosa Labs LLC, “AI Spend per Employee,” 2026-10-24.

The ledger

View

Denominator

Basis

Harmonisation

8 of 8 figures

#
1	Anthropic: Claude Code enterprise average per developerClaude Code documentation, Manage costs effectively [Primary verified]	?
2	Atlanta Fed SBU: top decile of US firms, expected AI spend 2026Survey of Business Uncertainty, March 2026 special module [Primary verified]	623.33USD
3	Atlanta Fed SBU: US firms, expected AI spend 2026Survey of Business Uncertainty, March 2026 special module [Primary verified]	170.25USD
4	Ramp AI Index: median firm paying for AI, July 2026Ramp AI Index [Primary verified]	11.95USD
5	Ramp AI Index: median firm paying for AI, September 2026Ramp AI Index [Primary verified]	15.7USD
6	Ramp AI Index: top 1% of firms paying for AI, July 2026Ramp AI Index [Primary verified]	7,400USD
7	Ramp AI Index: top 1% of firms paying for AI, September 2026Ramp AI Index [Primary verified]	9,567.53USD
8	Ramp AI Index: top 10% of firms paying for AI, September 2026Ramp AI Index [Primary verified]	813.43USD
		623.33(7/8)

AI Spend per Employee

Snapshot

2026-10-24

Coverage

8figures·3publishers·7on the common axis·5comparability groups

Download

Anthropic: Claude Code enterprise average per developerClaude Code documentation, Manage costs effectively

[Primary verified]

Atlanta Fed SBU: top decile of US firms, expected AI spend 2026Survey of Business Uncertainty, March 2026 special module

[Primary verified]

623.33USD

Atlanta Fed SBU: US firms, expected AI spend 2026Survey of Business Uncertainty, March 2026 special module

[Primary verified]

170.25USD

Ramp AI Index: median firm paying for AI, July 2026Ramp AI Index

[Primary verified]

11.95USD

Ramp AI Index: median firm paying for AI, September 2026Ramp AI Index

[Primary verified]

15.7USD

Ramp AI Index: top 1% of firms paying for AI, July 2026Ramp AI Index

[Primary verified]

7,400USD

Ramp AI Index: top 1% of firms paying for AI, September 2026Ramp AI Index

[Primary verified]

9,567.53USD

Ramp AI Index: top 10% of firms paying for AI, September 2026Ramp AI Index

[Primary verified]

813.43USD

623.33(7/8)