Skip to content

Research report ยท September 2026 edition

AI Training Market Concentration

On 30 Sep 2026, AI training listings were moderately concentrated overall (HHI 1,683 across 16 platforms); Red-Teaming & AI Safety was the most concentrated kind of work, led by Mercor with 81.6% of its open roles.

Read the month's story: The AI Training Job Market in September 2026

Data window
1 Sep 2026 to 30 Sep 2026
30 days, UTC
Frozen
10 Oct 2026, 10:32 UTC
These figures will not change
Source
The public job boards of the AI training platforms this site tracks, read every day.
Method
Version 1. Definitions
CC BY 4.0, how to cite

Every figure on this page as CSV and JSON, sent to your work email.

Key findings

  1. 1

    Across all AI training work, listings are moderately concentrated (HHI 1,683 across 16 platforms); OpenTrain AI holds 30.3% of open roles.

    Figure 1
  2. 2

    Red-Teaming & AI Safety is the most concentrated kind of work (HHI 6,911, Mercor with 81.6%); Transcription, Audio & Speech the least (HHI 1,689, across 10 platforms).

    Figure 1
  3. 3

    8 of 12 kinds of work with at least 10 open roles are highly concentrated by listings (HHI above 2,500).

    Figure 1
  4. 4

    On the same platforms, concentration fell from 1 Sep 2026 to 30 Sep 2026 (HHI 1,993 to 1,860) and fell in 4 of 11 kinds of work; the largest move was in Agentic Tasks & RL Environments, from 3,149 to 5,200.

    Figure 3
  5. 5

    OpenTrain AI lists the widest range of work: all 12 kinds of work with open roles.

    Figure 4

Figure 1

How concentrated each kind of work is

For which kinds of AI training work does one platform hold most of the openings?

Kind of workHHIReadingIts share
All AI training work
1,683
Moderately concentrated30.3%
Red-Teaming & AI Safety
6,911
Highly concentrated81.6%
Agentic Tasks & RL Environments
6,818
Highly concentrated81.5%
Research Studies & Surveys
5,156
Highly concentrated68.8%
Robotics & Physical-AI Data
4,412
Highly concentrated53.8%
Writing & Content
3,599
Highly concentrated53.7%
AI Evaluation & Rating (RLHF)
2,797
Highly concentrated44.1%
Coding & Software Evaluation
2,706
Highly concentrated42.2%
Data Annotation & Labeling
2,564
Highly concentrated40.9%
Math & STEM
2,498
Moderately concentrated34.1%
Law, Medicine & Finance
2,366
Moderately concentrated39.4%
Multilingual & Localization
1,842
Moderately concentrated32.0%
Transcription, Audio & Speech
1,689
Moderately concentrated25.7%

Sample. 2,729 roles open at the end of 30 Sep 2026 on the job boards this site reads, by the platform that lists them.

HHI: the sum of every platform's squared percentage share of the open roles, 0 (many small platforms) to 10,000 (one platform). Read with the US merger-guideline thresholds: under 1,500 unconcentrated, 1,500 to 2,500 moderately, above 2,500 highly concentrated.

Shares are of listings, not of people hired or money paid: a platform posting one role per country or language weighs more here than its hiring may.

Source. AITraining.jobs Research, AI Training Market Concentration, September 2026 edition.

Figure 2

The three largest platforms in each kind of work

Who leads each kind of AI training work, and how far ahead?

Kind of workFirstSecondTop three's share
Agentic Tasks & RL EnvironmentsOpenTrain AI (81.5%)Terac (11.1%)
100.0%
Red-Teaming & AI SafetyMercor (81.6%)OpenTrain AI (15.8%)
100.0%
Robotics & Physical-AI Datamicro1 (53.8%)OpenTrain AI (38.5%)
97.4%
Research Studies & SurveysTransPerfect DataForce (68.8%)Terac (18.8%)
93.8%
Writing & ContentOpenTrain AI (53.7%)micro1 (24.4%)
85.4%
AI Evaluation & Rating (RLHF)micro1 (44.1%)OpenTrain AI (22.0%)
84.1%
Math & STEMInvisible (34.1%)OpenTrain AI (26.4%)
83.6%
Coding & Software EvaluationOpenTrain AI (42.2%)micro1 (20.3%)
79.3%
Data Annotation & LabelingWelo Data (40.9%)OpenTrain AI (27.0%)
76.5%
Law, Medicine & FinanceOpenTrain AI (39.4%)AfterQuery (17.6%)
73.3%
Multilingual & LocalizationInvisible (32.0%)OpenTrain AI (20.8%)
66.8%
Transcription, Audio & SpeechOpenTrain AI (25.7%)Welo Data (22.1%)
63.1%

Sample. Kinds of work with at least 10 roles open on 30 Sep 2026.

Source. AITraining.jobs Research, AI Training Market Concentration, September 2026 edition.

Figure 3

Is each kind of work consolidating?

Did concentration rise or fall over the window, on the same platforms?

Kind of workHHI on 1 Sep 2026HHI on 30 Sep 2026Change
All AI training work1,993
1,860
-133
Red-Teaming & AI Safety8,828
9,395
+567
Robotics & Physical-AI Data8,893
9,132
+239
Research Studies & SurveysSmall sample3,333
5,822
+2,489
Agentic Tasks & RL Environments3,149
5,200
+2,051
Writing & Content3,979
4,375
+396
AI Evaluation & Rating (RLHF)4,482
4,191
-291
Math & STEM3,001
3,326
+325
Multilingual & Localization3,568
2,857
-711
Coding & Software Evaluation2,416
2,765
+349
Law, Medicine & Finance2,180
2,259
+79
Transcription, Audio & Speech2,479
1,984
-495
Data Annotation & Labeling2,378
1,614
-764

Sample. Roles open at the start and the end of the window on the platforms tracked throughout.

Platforms first tracked during the window are left out of both ends, so their arrival does not read as a change in competition.

Source. AITraining.jobs Research, AI Training Market Concentration, September 2026 edition.

Figure 4

Specialists and generalists

Which platforms spread across many kinds of work, and which specialise?

PlatformKinds of work listedShare in its main kind
OpenTrain AI
12
30.0%
AfterQuery
9
56.1%
Invisible
9
45.8%
micro1
9
31.0%
TransPerfect DataForce
9
40.9%
Mercor
8
20.0%
Braintrust
7
46.7%
Welo Data
7
43.3%
Alignerr
6
50.0%
Terac
6
33.3%
Appen
5
78.1%
xAI (SpaceXAI)
4
83.9%
RemotasksSmall sample
3
33.3%
Handshake AISmall sample
2
75.0%
Outlier (Scale AI)Small sample
2
66.7%
Sigma AI
2
90.8%

Sample. 2,729 roles open on 30 Sep 2026.

Specialisation uses the same index across a platform's own kinds of work: 10,000 means it lists one kind only.

Source. AITraining.jobs Research, AI Training Market Concentration, September 2026 edition.

For labs, data vendors and researchers

Buy the data behind this report

The figures above are a summary. Behind them are records we collect ourselves, every day:

  • 9,572 roles tracked, each with its history

    Every role on 29 job boards, read daily since 13 Jul 2026: its pay and place, and the first and last day it was open.

  • 6,200 apply clicks in 28 days

    Which roles people actually apply to, by field, pay and kind of work. It happens on this site, so nobody else can count it.

  • 39 lab and vendor hiring boards

    The job boards of 25 AI labs and 14 human-data vendors, read every day: who is hiring for human data, and who is hiring to buy it.

What you can buy

  • The records

    The rows behind any figure, with their history, not only the totals.

  • A regular feed

    The same figures every week or month, as files.

  • A custom cut or benchmark

    Your countries, languages or fields, or your own platform against the market.

Method and definitions

The public job boards of the AI training platforms this site tracks, read every day. Every figure is computed by one database query over the window shown, the same query for the live report and for every edition. Each figure states its sample beneath it.

HHI
Herfindahl-Hirschman Index: the sum of every platform's squared percentage share of open roles, from near 0 (many small platforms) to 10,000 (one platform). Read with the US merger-guideline thresholds: under 1,500 unconcentrated, 1,500 to 2,500 moderately concentrated, above 2,500 highly concentrated.
Share
A platform's open roles in a kind of work over all open roles in it, on the last day of the window.
Like for like
The change figure counts only platforms already tracked when the window opens, so a platform's arrival in this dataset is not read as a change in competition.
Open role
A role listed by a given day and still in its platform's feed that day.
Kind of work
Each role is classified from its title and description into one of the twelve kinds of AI training work this site tracks, from AI evaluation and coding to multilingual and audio work. Every kind links to its own page.
Coverage
The platforms whose public job boards AITraining.jobs reads in full every day. Each figure states how many were read.
Window
Live: the last 28 complete days, in UTC. Edition: one calendar month, in UTC.

What this data cannot tell you

  • Platforms that recruit without a public job board are not counted. Several large ones invite workers to projects directly.
  • A listing is not a hire. Some platforms keep evergreen roles open for months, and some post one listing per language or country.
  • Shares are of public listings, not of workers hired, hours worked or revenue; a platform posting one role per language or country weighs more here than its hiring may.

How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.

Editions

The live report covers the last 28 complete days and changes every day. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.

Cite and reuse

Free to quote, chart and republish under CC BY 4.0 with a link to this page. This edition is frozen, so its figures will read the same later.

Citation

AITraining.jobs (2026). AI Training Market Concentration, September 2026 edition. https://www.aitraining.team/research/ai-training-market-concentration/2026-09

BibTeX

@misc{aitrainingjobs_ai_training_market_concentration_2026_09,
  author = {{AITraining.jobs}},
  title = {AI Training Market Concentration, September 2026 edition},
  year = {2026},
  howpublished = {\url{https://www.aitraining.team/research/ai-training-market-concentration/2026-09}},
}
Data: every figure as CSV (one line per value) and JSON, sent to your work email.Pay and open-roles data

More research

All 78 research reports