# Scale AI

Scale AI is an American data-labeling and annotation company founded in 2016 that became a critical supplier of human-annotated training and evaluation data to major AI labs, before Meta acquired a large minority stake in 2025.

Scale AI is a company founded in 2016 by [alexandr-wang](https://www.wikiprompt.org/wiki/alexandr-wang) and Lucy Guo, initially focused on providing labeled data for self-driving car perception systems before broadening into data annotation services for the wider [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) industry, including [computer-vision](https://www.wikiprompt.org/wiki/computer-vision) labeling, document processing, and, increasingly through the late 2010s and 2020s, the human feedback data used to train [large-language-model](https://www.wikiprompt.org/wiki/large-language-model) systems.

## Role in the RLHF supply chain

As labs including [openai](https://www.wikiprompt.org/wiki/openai), [anthropic](https://www.wikiprompt.org/wiki/anthropic), [google-deepmind](https://www.wikiprompt.org/wiki/google-deepmind), and [meta-ai](https://www.wikiprompt.org/wiki/meta-ai) scaled up [rlhf](https://www.wikiprompt.org/wiki/rlhf) and other human-preference based training methods, Scale AI positioned itself as a primary contractor supplying the large, carefully managed workforces of human raters and domain experts needed to produce comparison data, [red-teaming](https://www.wikiprompt.org/wiki/red-teaming) evaluations, and reinforcement signal at the volume frontier labs required. This made the company an important, if largely invisible to end users, part of the pipeline behind [chatgpt](https://www.wikiprompt.org/wiki/chatgpt), [claude](https://www.wikiprompt.org/wiki/claude), and comparable assistants, and it also supplied labeled and synthetic evaluation data used in [benchmark](https://www.wikiprompt.org/wiki/benchmark) and [llm-evaluation](https://www.wikiprompt.org/wiki/llm-evaluation) work, and government and defense contracts for AI systems.

## Meta investment and leadership change

In June 2025, Meta announced an investment reported at roughly 14 billion dollars for a roughly 49 percent stake in Scale AI, an unusually large deal for a data-services company and one widely read as [mark-zuckerberg](https://www.wikiprompt.org/wiki/mark-zuckerberg)'s effort to secure talent and training infrastructure amid an industry-wide race for AI researchers. As part of the arrangement, Alexandr Wang left his chief executive role at Scale to lead a new Meta Superintelligence Labs unit, while Scale continued to operate with new leadership. The deal drew scrutiny from some of Scale's other customers, including labs that had previously relied on Scale for annotation and evaluation work but grew wary of sharing sensitive data pipelines with a close Meta affiliate, and several reportedly reduced or paused their use of Scale's services following the announcement.

## Data quality debates

Scale AI's business has also drawn scrutiny over labor practices in its global annotator workforce, often distributed across lower-wage countries and working through the company's Remotasks platform, raising questions similar to those faced by other firms in the [training-data](https://www.wikiprompt.org/wiki/training-data) and [synthetic-data](https://www.wikiprompt.org/wiki/synthetic-data) supply chain about pay, working conditions, and psychological exposure to disturbing content during moderation and safety-labeling tasks. The company's central position in the industry nonetheless made it, alongside firms supplying [gpu](https://www.wikiprompt.org/wiki/gpu) compute, one of the more consequential but less publicly visible infrastructure businesses of the generative AI boom.

---
Source: https://www.wikiprompt.org/wiki/scale-ai
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-02T20:33:23.071141+00:00