Skip to content
aiwork.fyi

The AI data industry

See who's getting rich on AI data — and what the people doing the work actually earn.

AI labs pay data vendors billions for human training data. Most of that money does not show up in what workers take home. This is a public record of the vendors — what they sell, what they report in revenue and valuation, and what they pay — with every figure still tied to its source.

Browse the vendorsReport your pay

Research as of Oct 7, 2026. 18 companies. Worker reports are published only after moderation, and only in groups of at least three.

Combined valuation

~$49B

up to ~$78B counting reported talks

Private AI data vendors with a priced or deal-implied valuation. Public companies and TELUS Digital are not included.

Combined run-rate

$6.7B+[est.]

a year, in what vendors bill AI labs

An estimate. Mixes gross and net figures from different dates, and excludes vendors that don't publish revenue.

Advertised vs. actual pay

Checking approved reports…

How this is counted

What we know

Three findings from the same research pass. Each claim keeps its source.

  1. 01

    Platforms keep roughly 30–40% of gross billings: Handshake ~$300M net on ~$1B gross (Dealroom); micro1 $150–200M net on $500M (TechCrunch); Prolific participants get ~70% of spend (Prolific).

  2. 02

    In 2023, OpenAI reportedly paid Sama $12.50/hr per worker while Kenyan workers took home $1.32–2/hr (TIME).

  3. 03

    Google super raters earned $28–32/hr direct vs $18–22 through subcontractors (WIRED).

Largest headline revenues

Ordered by the latest figure named in the sourced text. The sentence is the record, including when a source disagrees with itself.

Full table
  1. 1Mercor

    >$2B gross run-rate (Jun 2026); H1 2026 gross ~$614M; net est. $600–800M

  2. 2Surge AI (DataAnnotation.tech)

    2024: $1.2B [reported]; 2025: $1.4B [unverified, aggregator]

  3. 3Handshake AI

    ~$1B gross AI-training run-rate (Apr 2026), ~$300M net; $550M in Jan 2026

  4. 4Scale AI (Outlier, Remotasks)

    2025: just shy of $1B [company via Forbes, May 2026]; 2024: $870M. Sacra est. $2B for 2025 (conflicts)

  5. 5micro1

    $500M gross run-rate (Aug 2026), net $150–200M; $100M ARR in Dec 2025

  6. 6Prolific

    ~$350M annualized (Apr 2026) [est., Sacra]

  7. 7Turing

    $300M+ ARR (end 2024, profitable) [company]; 2025–26 not public

  8. 8Appen (ASX: APX)

    FY25: US$230.8M, net loss US$21.8M; FY26 guidance $270–300M [company]

Not ranked here, because the public number is not a comparable AI-training figure: TELUS Digital, Toloka (Mindrift), Sama, GlobalLogic (Hitachi).

  • [reported] named source, not audited here
  • [unverified] not tied to a filing or the company
  • [est.] an estimate
  • [company] the company's own figure
  • Not public — no public figure in this pass

How a pay report becomes a number

  1. 1. Submit what you earned.

    Company, project type, country, the advertised rate, and the effective hourly rate. No name. Email is optional and never displayed.

  2. 2. The report stays pending until it is reviewed.

    Only approved reports can count. Notes are published only after that review.

  3. 3. Three approved reports publish a median.

    The headline number is the median effective hourly rate for that company and project type, next to the gap versus the advertised rate.

Report your pay

All 18 companies