Skip to content

Best AI Data Labs in 2026: 38 Training Data Companies Ranked

Bar chart: the 10 biggest AI data labs by revenue in 2026. Mercor $2B+ run-rate, Surge AI $1.2B, micro1 $1B+, Handshake AI $1B+, Scale AI just under $1B, Snorkel AI $375M+, Innodata $350M, Turing $300M, TaskUs AI Services $261M, Appen $249M.

The biggest AI data labs in 2026 are Mercor, with a $2B+ revenue run-rate, Surge AI, which made $1.2B in 2024 without raising outside money, and micro1, which says it passed a $1B run-rate in October. An AI data lab is a company whose main business is making the data that trains and tests AI models: expert answers, rankings, evaluations and practice environments. This list ranks 38 of them by revenue, then valuation.

One caution before the numbers: several of these companies report gross revenue, which includes the 60–70% they pay out to the experts doing the work. We say which figures are gross.

Top 10 AI data labs in 2026, ranked

#

Company

Kind

HQ

Revenue

Valuation

Main customers

1

Mercor

Expert data

San Francisco, US

$2B+ run-rate, gross (Jun 2026)

$20B (Nov 2026)

OpenAI, Anthropic, Meta

2

Surge AI

Expert data

San Francisco, US

$1.2B (2024, reported)

no outside funding

Google, OpenAI, Anthropic

3

micro1

Expert data

San Francisco, US

$1B+ run-rate (Oct 2026)

$4B (Sep 2026)

Microsoft, Amazon

4

Handshake AI

Expert data

San Francisco, US

$1B+ run-rate, gross (Aug 2026, reported)

$3.5B (2022, before the AI unit)

OpenAI and 7 other labs

5

Scale AI

Expert data

San Francisco, US

just under $1B (2025, reported)

$29B (Jun 2025, Meta deal)

Meta, Cisco, enterprises

6

Snorkel AI

Expert data

San Francisco, US

$375M+ run-rate (Sep 2026)

$3.5B (Sep 2026)

Google, Anthropic, OpenAI

7

Innodata

Annotation workforce

Ridgefield Park, US

$350M (12 months to Sep 2026)

$1.1B market cap (Nov 2026)

5 of the Magnificent Seven

8

Turing

Expert data

San Francisco, US

$300M run-rate (end of 2024)

$2.2B (Mar 2025)

OpenAI

9

TaskUs

Annotation workforce

New Braunfels, US

$261M, AI Services unit (12 months to Sep 2026)

$846M market cap, whole company (Nov 2026)

Meta

10

Appen

Annotation workforce

Sydney, Australia

$249M (12 months to Jun 2026)

$192M market cap (Nov 2026)

Amazon, Microsoft, Nvidia

Key numbers

  • Mercor is the biggest by revenue: a $2B+ gross run-rate in June 2026, and a $20B valuation after its November Series D. About 91% of its gross revenue in the first half of 2026 reportedly came from OpenAI and Anthropic.

  • Scale AI is the most valuable: Meta paid $14.3B for 49% of it in June 2025, valuing Scale at about $29B. Google and OpenAI then moved work to its rivals.

  • Surge AI never raised money and still made a reported $1.2B in 2024, more than Scale's $870M.

  • At least 12 of the 38 companies were founded in 2022 or later. The youngest grow fastest: Proximal says it passed a $200M run-rate about 10 months after starting.

  • Expert pay is the business. Mercor's homepage claims a $90/hour average rate and $4M+ paid to experts every day.

  • The field is crowded: Pavlov’s List tracks 71 startups selling RL environments and training data (October 2026).

Bar chart: latest valuations of AI data labs. Scale AI $29B (Meta deal, June 2025), Mercor $20B (November 2026), micro1 $4B, Snorkel AI $3.5B, AfterQuery $3.2B (reported), Turing $2.2B, Invisible $2B+, Labelbox $1.25B pre-money, Innodata $1.1B market cap.

What is an AI data lab?

An AI data lab makes data to train or evaluate AI models and sells it to the companies that build them. The work ranges from simple labeling (drawing boxes on images, tagging text) to paying doctors, lawyers and engineers to write and grade answers, and building simulated software where AI agents can practice.

It's not the same as an AI research lab. A research lab like OpenAI or Anthropic trains the models; a data lab supplies them. The model builders are in our list of AI labs, and the models themselves in our ranking of the best LLMs. We left out web-data APIs like Firecrawl, data warehouses and synthetic-data tools, which are in our lists of AI infrastructure companies and AI inference companies.

The five kinds of AI data lab

1. Expert human data for frontier labs

The biggest and fastest-growing group. These companies recruit experts, from coders to physicians, to write example answers, rank model outputs (RLHF) and grade them against rubrics. Most of their money comes from a handful of frontier labs. In this list: Mercor, Surge AI, micro1, Handshake AI, Scale AI, Snorkel AI, Turing, Invisible Technologies, AfterQuery, Prolific, Deccan AI, Datacurve, Toloka, Taste Labs.

2. RL environments and evaluations

The newest group. An RL environment is a simulated copy of real software, like Salesforce, Slack or a code base, where an AI agent practices tasks and gets scored. Epoch AI estimates that tasks typically cost $200–$2,000 each, a copy of a website about $20,000, and a high-fidelity copy of an app like Slack around $300,000. Anthropic reportedly discussed spending more than $1B on RL environments over a year. In this list: Proximal, Fleet, Mechanize, Patronus AI, Bespoke Labs, Collinear AI. Mercor bought Deeptune, another environment builder, in 2026.

3. Labeling platforms

Software for labeling and checking data, sold to AI teams that do the work themselves. Most now also sell the data: Labelbox through its Alignerr expert network, HumanSignal through HumanSignal Services. The group is consolidating: Dell agreed to buy Dataloop (reported), and Uber bought Segments.ai. In this list: Labelbox, Encord, SuperAnnotate, Kili Technology, HumanSignal.

4. Managed annotation workforces

The older model: large teams, often in Asia or Africa, that label data at volume, plus crowds of freelancers. Three are listed on stock exchanges, so their revenue is the most reliable on this page. Sama, a pioneer of the model, is winding down. In this list: Innodata, TaskUs, Appen, iMerit, Shaip, Defined.ai, CloudFactory, Cogito Tech, LXT.

5. Robotics and physical-world data

Data for robots: people filmed doing everyday tasks with head-mounted cameras, motion-capture suits and teleoperated robot arms. Mecka AI says its run-rate passed $100M in June 2026, a year after it started. In this list: Mecka AI, XDOF, Maxinsights, Ropedia. Encord, Innodata and micro1 also sell robotics data.

Mecka AI homepage: the data and deployment layer for physical AI, with a note on its $60M Series B.

All 38 AI data labs compared

Ranked by revenue where it's public, then by valuation, then by headcount, then by total funding. CEO names only; for their contact details, see the call to action below.

#

Company

Kind

HQ

Founded

Employees

Funding or valuation

CEO

1

Mercor

Expert data

San Francisco, US

2023

—

$20B (Nov 2026)

Brendan Foody

2

Surge AI

Expert data

San Francisco, US

2020

267 (2026)

no outside funding

Edwin Chen

3

micro1

Expert data

San Francisco, US

2022

180 (2026)

$4B (Sep 2026)

Ali Ansari

4

Handshake AI

Expert data

San Francisco, US

2014

—

$3.5B (2022, before the AI unit)

Garrett Lord

5

Scale AI

Expert data

San Francisco, US

2016

1,300 (2026)

$29B (Jun 2025, Meta deal)

Francis deSouza

6

Snorkel AI

Expert data

San Francisco, US

2019

125 (2026)

$3.5B (Sep 2026)

Alex Ratner

7

Innodata

Annotation workforce

Ridgefield Park, US

1988

10,107 (2025)

$1.1B market cap (Nov 2026)

Rahul Singhal

8

Turing

Expert data

San Francisco, US

2018

—

$2.2B (Mar 2025)

Jonathan Siddharth

9

TaskUs

Annotation workforce

New Braunfels, US

2008

64,700 (whole company)

$846M market cap, whole company (Nov 2026)

Bryce Maddock

10

Appen

Annotation workforce

Sydney, Australia

1996

—

$192M market cap (Nov 2026)

Ryan Kolln

11

Proximal

RL environments and evals

San Francisco, US

2025

40 (2026)

$300M (seed, 2025)

Calvin Chen

12

Invisible Technologies

Expert data

San Francisco, US

2015

350 (2025)

$2B+ (Sep 2025, reported)

Matthew Fitzpatrick

13

Mecka AI

Robotics data

New York, US

2025

45 (2026)

~$500M (Sep 2026, reported)

Josh Gao

14

AfterQuery

Expert data

San Francisco, US

2025

30 (2026)

$3.2B (Sep 2026, reported)

Spencer Mateega

15

Fleet

RL environments and evals

San Francisco, US

—

—

$750M (Apr 2026, reported)

Nicolai Ouporov

16

Prolific

Expert data

London, UK

2014

162 (2025)

raised $32M

Phelim Bradley

17

Labelbox

Labeling platform

San Francisco, US

2018

141 (2026)

$1.25B pre-money (Oct 2026)

Manu Sharma

18

Encord

Labeling platform

San Francisco, US

2020

150 (2026)

$550M (Feb 2026, reported)

Eric Landau and Ulrik Stig Hansen

19

Mechanize

RL environments and evals

San Francisco, US

2025

35 (2026)

$500M (2026, reported)

Guive Assadi

20

iMerit

Annotation workforce

Kolkata, India

2012

3,000 (2020)

bought by EXL for up to $310M (Aug 2026)

Radha Basu (head of iMerit at EXL)

21

Shaip

Annotation workforce

New York, US

2019

150 (2026)

—

Matt Nyren (Ubiquity)

22

Deccan AI

Expert data

SF Bay Area, US

2024

125 (2026)

raised $25M

Rukesh Reddy

23

SuperAnnotate

Labeling platform

San Francisco, US

2018

100 (2024)

raised $66.5M

Vahan Petrosyan

24

XDOF

Robotics data

US

2024

60 (2026)

raised $70M

Philipp Wu

25

Kili Technology

Labeling platform

Paris, France

2018

43 (2026)

raised $30M

François-Xavier Leduc

26

Datacurve

Expert data

San Francisco, US

2024

25 (2026)

raised $17.7M

Serena Ge

27

Defined.ai

Annotation workforce

Seattle, US

2015

—

raised $85M

Daniela Braga

28

CloudFactory

Annotation workforce

Reading, UK

2010

—

raised $78M

Kevin Johnston

29

Toloka

Expert data

Amsterdam, Netherlands

2014

—

raised $72M

Olga Megorskaya

30

Patronus AI

RL environments and evals

San Francisco, US

2023

—

raised $70M

Anand Kannappan

31

Bespoke Labs

RL environments and evals

SF Bay Area, US

—

—

raised $40M

Mahesh Sathiamoorthy

32

HumanSignal

Labeling platform

San Francisco, US

2019

—

raised $30M

Michael Malyuk

33

Maxinsights

Robotics data

Santa Clara, US

—

—

raised $30M

Yuguang (YG) Yuan

34

Ropedia

Robotics data

Singapore and Mountain View, US

2025

—

raised $30M

Zhaoxi Chen

35

Taste Labs

Expert data

San Francisco, US

2025

—

raised $18.5M

Thais Castello Branco

36

Cogito Tech

Annotation workforce

Levittown, US

2011

—

—

Rohan Agrawal

37

Collinear AI

RL environments and evals

Sunnyvale, US

—

—

—

Nazneen Rajani

38

LXT

Annotation workforce

Toronto, Canada

2010

—

—

Mohammad Omar

The top 10 in detail

1. Mercor

Mercor, founded in 2023, recruits and pays domain experts, from doctors and lawyers to bankers and engineers, to write training data, evaluations and RL environments for AI labs.

  • Revenue: a $2B+ gross run-rate in June 2026, said CEO Brendan Foody. Experts get 60–70% of it.

  • Valuation: $20B after a $750M Series D on November 10, 2026, double its $10B valuation of October 2025.

  • Customers: OpenAI, Anthropic and Meta. About 91% of its first-half 2026 gross revenue reportedly came from OpenAI and Anthropic.

  • 2026: bought Sepal AI (February) and agreed to buy the RL-environment builder Deeptune (July).

Mercor homepage: organizing human intelligence to power the AI economy, with links to find work, evaluate AI, build evals and sell data.

2. Surge AI

Surge AI sells high-skill human data, such as preference rankings, expert-written examples and evaluations, to frontier labs. It was founded in 2020 by Edwin Chen and has never raised outside money.

  • Revenue: $1.2B in 2024 (reported), more than Scale AI's $870M that year. It hasn't published a newer figure.

  • Valuation: none. In July 2025 it was reportedly in talks for a first raise at $25B or more, but by September 2026 it still hadn't taken outside money.

  • Customers: Google, OpenAI and Anthropic.

  • Contributors: it hires through DataAnnotation, which a 2025 lawsuit names as a Surge business (“Surge Labs, Inc. dba DataAnnotation”).

Surge AI homepage: “Our mission is to raise AGI with the richness of humanity: curious, witty, creative, and wise.”

3. micro1

micro1 sells expert human data, RL environments and evaluations to frontier labs and enterprises, plus real-world data for robotics. Its core team is about 180 people, fully remote.

  • Revenue: passed a $1B run-rate in October 2026, the company says, while “historically profitable”. That's its own unaudited figure; TechCrunch put its gross run-rate at $500M in August.

  • Valuation: $4B in a Series B of more than $100M, eight times its valuation a year earlier.

  • Customers: Microsoft, Amazon and the robotics company 1X. CEO: Ali Ansari.

micro1 homepage: data lab to train frontier models and evaluate agents.

4. Handshake AI

Handshake, the campus recruiting network, turned its network of students, graduates and PhDs into a source of expert training data. Its AI unit is now far bigger than the recruiting business.

  • Revenue: the AI unit reportedly went “from zero to more than $1 billion in gross annualized revenue in 16 months” (August 2026).

  • Valuation: $3.5B in 2022, before the AI unit existed. No newer figure is public.

  • Customers: eight top AI labs, including OpenAI. In January 2026 it bought Cleanlab, a data-quality startup. CEO: Garrett Lord.

5. Scale AI

Scale AI, founded in 2016, is the best-known name in data labeling. It now sells training and evaluation data to AI labs, plus custom AI systems for companies and governments.

  • Revenue: just under $1B in 2025 (reported), up from $870M in 2024; Scale expects more than $1B in 2026.

  • Valuation: about $29B, from Meta's $14.3B investment for 49% in June 2025. Founder Alexandr Wang moved to Meta.

  • After the deal: OpenAI phased out its work with Scale, Scale cut 14% of staff in July 2025, and Francis deSouza became CEO on August 10, 2026.

  • Contributors: Outlier, which says it is “a platform operated by Scale AI”.

6. Snorkel AI

Snorkel AI started as labeling software and now sells finished datasets, RL environments and evaluations built by experts with its own tools, which it calls “data-as-a-service”.

  • Revenue: a $375M+ run-rate in September 2026, up more than 18x in under a year. Snorkel counts expert pay as a cost, so this is closer to net revenue than Mercor's figure.

  • Valuation: $3.5B after a $350M Series E, about triple its valuation 17 months earlier.

  • Customers: its site shows Google, Anthropic, OpenAI, Microsoft, AWS and Mistral AI. CEO: Alex Ratner.

Snorkel AI homepage: “We build the data that pushes the frontier.”

7. Innodata

Innodata, listed on Nasdaq and founded in 1988, makes training, fine-tuning and evaluation data for big tech's AI models, with about 10,000 staff, mostly in Asia.

  • Revenue: $350M in the 12 months to September 2026, and a record $95.6M in the third quarter, up 53%.

  • Customers: five of the “Magnificent Seven” tech giants. One customer was about 58% of its 2025 revenue.

  • 2026: opened a motion-capture lab for humanoid-robot data, and Rahul Singhal took over as CEO.

8. Turing

Turing supplies coding, reasoning and STEM training data and RL environments to AI labs, drawing on a network of about 4 million developers, and builds AI systems for enterprises.

  • Revenue: a $300M run-rate at the end of 2024, in a profitable year. It hasn't published a newer figure.

  • Valuation: $2.2B in a $111M Series E in March 2025.

  • Customers: OpenAI, among others. CEO: Jonathan Siddharth.

9. TaskUs (AI Services)

TaskUs is an outsourcing company listed on Nasdaq. Only its AI Services unit, which does LLM support and data labeling for AI developers, counts here; most of the company does customer support and content moderation.

  • Revenue: $261M from AI Services in the 12 months to September 2026, out of $1.24B in total.

  • Customers: Meta was 26% of its 2025 revenue.

  • 2025: a take-private by its founders and Blackstone failed in a shareholder vote, so it stays listed.

10. Appen

Appen, founded in Sydney in 1996 and listed on the ASX, is the veteran of the group: it supplies human-labeled training and evaluation data through a crowd of more than 1 million contractors.

  • Revenue: $249M in the 12 months to June 2026, with $270–300M guided for 2026. It lost its Google contract in 2024.

  • Growth: Appen China grew 80% in the first half of 2026, to a run-rate above $175M, while the rest of Appen shrank.

  • Contributors: CrowdGen, “the contributor platform of Appen”. CEO: Ryan Kolln.

Who buys AI training data

A handful of frontier labs and big tech companies buy most of it: OpenAI, Anthropic, Google, Meta, Microsoft, Amazon and xAI. That makes the data labs' revenue concentrated and fragile.

  • Before Meta's deal, Google reportedly planned to pay Scale AI about $200M in 2025, after about $150M in 2024, then planned to split.

  • OpenAI said it left Scale because it “needed more data expertise”. Meta's own new lab works with Scale's rivals Mercor and Surge.

  • xAI cut about 500 generalist annotators and said it would grow its team of specialist tutors tenfold.

  • Chinese labs are buyers too: AfterQuery reportedly gets at least $50M of its recurring revenue from them.

  • Labs also build data in-house: Epoch AI found they are doing “substantially more in-housing” of RL-environment work.

The shift is clear: from cheap volume labeling to expensive expert data and RL environments. Selling to these buyers? See the call to action below.

Working for an AI data lab

Most of the work is done by remote contractors, not employees, through each lab's own contributor platform. Pay depends on your field and the project, and the rates below are what each platform publishes, not a promise.

Platform

Run by

Published pay

Apply

Outlier

Scale AI

Not published; set per project, paid weekly

outlier.ai

DataAnnotation

Surge AI

Up to $150+/hour for coding, $125+ for experts, $50 for generalists

www.dataannotation.tech

Mercor

Mercor

$90/hour average contracted rate

work.mercor.com/explore

Handshake AI Fellowship

Handshake

$17–150/hour in its listings; US only

joinhandshake.com/fellowship-program

Alignerr

Labelbox

$80/hour average; roles $40–120/hour

www.alignerr.com

Mindrift

Toloka

$15–30/hour entry level; $60–100+ for experts

mindrift.ai

CrowdGen

Appen

Up to $450/hour for clinicians, $400 for legal

crowdgen.com

Meridial

Invisible Technologies

Set per project; a sample US posting paid $30–50/hour

www.meridial.ai/projects

micro1

micro1

Not published

www.micro1.ai/experts

Prolific

Prolific

At least £6/$8 per hour for research studies

www.prolific.com/participants

Handshake AI Fellowship page: “Get paid to advance AI”, with listings from $17 to $150 per hour.

The downsides, all on the record:

  • Surge AI and Scale AI have both been sued by contributors who say they were misclassified as contractors. These are allegations.

  • Work can stop overnight: Scale cut ties with 500 contractors in July 2025, and xAI let 500 annotators go in September 2025.

  • Rates can drop: Mercor reportedly replaced a Meta project with a similar one paying $5 an hour less, a report Mercor called inaccurate.

What changed in 2025 and 2026

  • June 2025: Meta bought 49% of Scale AI for $14.3B and hired its CEO. Google, OpenAI and others moved work to Scale's rivals.

  • 2025–2026: labs switched from generalist labelers to experts. Mercor, Surge, micro1 and Handshake AI passed $1B in revenue or run-rate.

  • September 2025 onward: RL environments became a category of their own, with Proximal, Fleet, Mechanize and others selling them.

  • 2026: consolidation. Handshake bought Cleanlab, Mercor bought Sepal AI and Deeptune, EXL bought iMerit, Ubiquity bought Shaip and Uber bought Segments.ai.

  • Fall 2026: record valuations: Snorkel AI at $3.5B (September), micro1 at $4B (September) and Mercor at $20B (November).

  • October 2026: Sama, an early leader in volume labeling, began winding down.

FAQ

What is the best AI data lab?

By revenue, Mercor ($2B+ gross run-rate), followed by Surge AI, micro1, Handshake AI and Scale AI. The best fit depends on the job: expert data (Surge AI, Mercor), RL environments (Proximal, Fleet), labeling software (Labelbox, Encord) or robotics data (Mecka AI).

Which companies label data for OpenAI?

Named in public reports: Mercor, Surge AI, Handshake AI and Turing. OpenAI phased out its work with Scale AI in 2025.

Is Scale AI owned by Meta?

Not fully. Meta bought 49% of Scale AI for $14.3B in June 2025, and Scale's founder Alexandr Wang moved to Meta. Scale is still run as its own company, led by CEO Francis deSouza since August 2026.

How much does Surge AI make?

Surge AI reportedly made $1.2B in revenue in 2024, without raising outside money. It hasn't published a newer figure.

What is RLHF data?

RLHF stands for reinforcement learning from human feedback. People compare or rate a model's answers, and the model learns from those preferences. It's the main product of expert data labs like Surge AI and Mercor.

What is an RL environment?

A simulated copy of real software, such as Salesforce or Slack, where an AI agent practices tasks and gets a score. Labs pay $200–$2,000 per task and up to about $300,000 for a full copy of an app, Epoch AI estimates.

Can I work for an AI data lab from home?

Yes. Outlier (Scale AI), DataAnnotation (Surge AI), Mercor, Alignerr (Labelbox), Mindrift (Toloka) and the Handshake AI Fellowship all hire remote contractors. Pay is set per project, and work can end without notice.

How many AI data labs are there?

This list covers 38 that sell mainly to AI model builders. Pavlov's List tracks 71 startups selling RL environments and training data, and there are hundreds of smaller labeling firms.

How this list was built

The list: companies whose main business is making data to train or evaluate AI models, in five groups: expert data, RL environments and evaluations, labeling platforms, annotation workforces and robotics data. We started from the companies tagged “AI Dataset” in LeadWave's database, added the ones that press coverage and the companies' own customer lists point to, and checked each one.

The ranking: companies with a public revenue figure come first, ranked by their latest annual revenue or run-rate. Then companies with no public revenue, by their latest valuation; then by headcount; then by total funding. Gross figures, which include what's paid to experts, are marked. A run-rate is a recent month or quarter multiplied out to a year.

The facts: checked on November 17, 2026, against company announcements and filings (Innodata and TaskUs with the SEC, Appen with the ASX, Prolific at Companies House) first, then the business and tech press. Figures from the press are marked “reported”. Headcounts come from the companies or LeadWave's database.

Left out: companies bought and folded into the buyer (Sepal AI and Deeptune into Mercor, Dataloop into Dell, Segments.ai into Uber); Sama, which is winding down; TELUS Digital, which doesn't report its AI data revenue; companies that pivoted away from data (V7, Superb AI, Datasaur and others); and model builders, security tools and compute platforms.

Selling to AI companies?

LeadWave's list of AI companies covers 9,000 AI companies with CEO details, including the data labs on this page and the labs that buy from them. For the CEOs' emails, see our AI companies email list. See also our list of AI labs and the best LLMs.

About this list

Compiled by LeadWave from company announcements, filings and press coverage, checked November 17, 2026.