Scale AI

basicwidespread

Scale AI is an American company that supplies human-labeled and verified training data to companies and government agencies developing artificial intelligence.

In plain terms

An AI model learns from what is shown to it. For a self-driving car to recognize a pedestrian, someone must first mark where the pedestrian is on millions of images. Scale AI organizes this background work at an industrial scale.

It is like a textbook publisher and a teaching staff combined: the student (the model) sits elsewhere, but Scale AI provides the curriculum and the corrected papers. The company is not known for its own chatbot, but for providing high-quality data to other people's models.

In the era of language models, the work has changed: it is no longer just about drawing boxes around objects in images; doctors, lawyers and programmers now write sample answers and evaluate models' responses. The company also does model testing and works on government and defense contracts.

How it works

The client (for example, an AI lab or car manufacturer) provides raw data or specifies what tasks examples should be created for.
Scale AI's software breaks the work down into small tasks and pre-labels whatever can be done automatically.
Tasks are performed by contract remote workers from many countries on the company's platforms (Remotasks, Outlier): they label, write responses, and score model answers.
Multi-stage quality control follows: multiple people review the same task, and faulty work is filtered out.
The client receives the finished, verified data package, which they use to train or test their own model.

An everyday example

When a chatbot gives a surprisingly precise answer to a tax or medical question, it is likely that experts previously wrote and scored similar answers for its training. Scale AI organizes this work, and you may encounter its advertisements, mostly under the name Outlier, if you are looking for a remote "AI trainer" job.

Who is affected

developerIf you are training or evaluating a model, Scale AI is one of the best-known vendors from which you can order data and testing. → Code

jobseeker, new graduateThe company advertises remote, task-paid data labeling and AI trainer jobs on its platforms, with varying conditions. → Work & HR

entrepreneur, executiveIf you have your own AI system built, you must decide whom to entrust your data to and how independent your supplier is from your competitors.

investorMeta's 2025 investment of more than 14 billion dollars demonstrated how much value the market sees in training data.

A common misconception

MythArtificial intelligence learns on its own, without human work.

In realityBehind the models lies extensive human work in meticulous labeling and evaluation. Scale AI's business is precisely organizing this invisible work.

Limits and risks

Most labelers are contract remote workers; the company has faced multiple criticisms and lawsuits due to low fees and uncertain payments.
Since Meta's 49 percent non-voting stake, the company's neutrality has been questioned: media reports say several major AI labs (including OpenAI and Google) have scaled back or halted cooperation for this reason.
The model is only as good as the data: faulty or biased labeling silently becomes part of the finished system.
The company also works on defense contracts, which raises ethical debates about the military use of AI.

Similar, but not the same

AI lab (e.g., OpenAI, Anthropic) – The lab develops the model itself, while Scale AI mainly supplies the necessary data and evaluation.
Data labeling – Data labeling is the activity itself, while Scale AI is a company that provides this as a service.
Meta – Meta acquired a 49 percent non-voting minority stake in Scale AI, but did not acquire or merge it: Scale AI continues to operate as a separate company.

Related concepts

Broader

AI company, data provider, AI supply chain

Narrower

Remotasks, Outlier

On the edge

Data labeling, Training data, Model evaluation, Meta, Training based on human feedback

A brief history

The company was founded in San Francisco in 2016 by Alexandr Wang and Lucy Guo. It started out as a general service brokering human work (under the name Scale API), but autonomous-vehicle developers soon became its main clients, for whom it labeled images and sensor data. In June 2025, Meta acquired a 49 percent non-voting stake in it for about 14.3 billion US dollars. Wang then moved to Meta, where he became chief AI officer and took charge of the newly created superintelligence division; Jason Droege became interim CEO of Scale AI.

Where things stand

Scale AI remains one of the best-known names in the training data market, but after Meta's entry, several major clients left, and it cut staff in 2025. Its focus has shifted toward expert data work, model evaluation, and government contracts.

Status: 2026 November

In the news

Appears in 62 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Microsoft scales back neocloud deals: winds down Oracle rental, turns down NscaleNov 4, 2026 · Agent Age
Siemens speeds up industrial AI scale-up and China localization, presents Eigen engineering agentNov 4, 2026 · Ipar
South Korea launches AI data factory with 12 bipedal humanoid robotsNov 3, 2026 · Ipar
Foxconn Technology Day: from manufacturer to AI factory player through vertical integrationNov 2, 2026 · Ipar
noyb files five complaints against Scale AI over surveillance of European AI trainersNov 2, 2026 · Jog
Scale AI and Missouri partners offer free AI training for teachersOct 30, 2026 · Education & Family
Acquisition and investment boost AI-based transaction monitoring and AML testingOct 28, 2026 · Money
Rockwell: food and beverage manufacturers struggle to scale AI beyond pilotsOct 28, 2026 · Ipar

Also known as

Scale AI · Scale · Scale AI, Inc. · ScaleAI · Scale.ai · Scale API · Scale.com

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.