Scale AI: Data Labeling, Evaluations & Meta's $14.3B Stake

The layer underneath the model

Scale AI was founded in San Francisco in 2016 by Alexandr Wang and Lucy Guo, who had worked together at Quora; Guo left the company in 2018. Its business is the unglamorous work that sits underneath a trained model: labeling the data models are trained on, running reinforcement learning from human feedback, and building the evaluations labs use to measure whether a new model actually improved. The company says its data has powered the majority of leading generative model builders, a self-reported figure rather than an independently audited one. [5][1]

Government work is a specific and growing part of that business. Scale AI holds contracts with the Department of Defense's Chief Digital and Artificial Intelligence Office, including a 2022 agreement reported at $250 million, and markets a dedicated national-security product line. Two subsidiaries, Remotasks and Outlier, run the human labeling and annotation workforce behind the labeling business, covering areas from computer vision to language model training data. [1][5]

Growing alongside the labs it serves

Scale AI reached unicorn status in 2019 after a $100 million investment from Peter Thiel's Founders Fund. Its May 2024 Series F raised $1 billion at a $13.8 billion valuation, led by Accel with Amazon and Meta among the participants, on the wager that frontier labs would keep paying for labeled data regardless of who ended up ahead in the model race. That wager, and Meta's participation specifically, foreshadowed the much larger deal that followed a year later. [6]

Meta buys a stake and a chief AI officer

In June 2025, Meta invested $14.3 billion for 49 percent of Scale AI, a non-voting stake that left Scale operating as an independent company. Alexandr Wang left to lead a new superintelligence lab inside Meta, and Jason Droege, previously the company's chief strategy officer, became interim chief executive. The New York Times later reported that lab was already discussing major strategy changes within weeks of Wang's arrival. [7][8]

Scale AI has since named Francis deSouza as chief executive, according to the company's own site, succeeding Droege's interim tenure. [5]

Atlas interpretation: The deal is a useful marker for where the industry's scarce resource had moved. Meta did not buy Scale AI outright; it bought a large minority stake and, more importantly to Meta, its founder and a team to run a new research effort. Scale's data and evaluation business kept operating for its existing customers, several of them Meta's own competitors, which is the arrangement Meta's 49 percent, non-voting structure was built to allow. [7]