Understand any stock in simple English
Enter a ticker to learn what the company does and how it makes money.
Scale AI organizes and labels data — like photos, text, and video — so that artificial intelligence models can learn from it.
Scale AI is a data company that helps artificial intelligence systems learn. Its core product is labeled data: humans, often contractors around the world, look at images, text, audio, or video and mark them up so a machine-learning model can understand what it is seeing. A self-driving car team, for example, needs thousands of street photos with every car, pedestrian, and stop sign outlined. Scale provides that work, plus software to manage it at large scale.
The business started in 2016 with autonomous driving, where the appetite for labeled sensor data was enormous. Scale's founder, Alexandr Wang, dropped out of MIT at 19 to build it. As the AI boom shifted from cars to chatbots, Scale moved with it. Today a big part of its business is "human feedback" — paying people to compare, rate, and correct AI answers so models like large language models behave better. It also runs a platform called Scale GenAI that lets companies test and evaluate their AI systems, and it offers services for governments and the military, which has become a large and controversial customer segment. Scale makes money by charging customers for the data work and software, usually through contracts that grow as a customer's AI projects grow. Its customers include big AI labs and tech companies, and Scale has worked with OpenAI, Meta Platforms, and the U.S. Department of Defense, among others. The appeal is speed and scale: a customer would otherwise have to hire and manage thousands of annotators itself. Competitors include other data-labeling firms and cheaper crowdsourcing platforms, and the rise of synthetic data — AI-generated training data — is a real threat to the human-labeling model. Scale's bet is that human judgment stays essential for the hardest, most valuable tasks.
Alexandr Wang and Lucy Guo met at Quora, tried a few ideas that went nowhere, and joined Y Combinator in 2016. Their bet was simple and strange at the time: the stuff that trains AI models, the data, matters as much as the models themselves. So they built a company to label data for AI, an unglamorous job nobody had turned into a real business yet. The first customers were self-driving car companies that needed images and laser scans tagged so their cars could learn to see.
Two years in, the partnership cracked. Guo and Wang disagreed over where the product should go, and it got ugly enough that she told someone Wang should be fired. When it all came out, she felt betrayed, and she left. Wang stayed on as the sole leader, and he later shrugged it off as a normal startup breakup.
Peter Thiel's Founders Fund put in a hundred million dollars, pushing Scale past a billion-dollar valuation. It was a big stamp of approval for a business that boiled down to people tagging pictures for computers, and self-driving demand kept growing.
Scale started doing work for the Pentagon, which was a controversial choice back when defense tech was not a cool place to be. The bet was that the stakes were too high to sit out. It paid off: government and defense became a steady pillar of the business.
Scale won a $250 million contract opening its tools to American federal agencies. Government work kept deepening, and within a couple of years Scale had earned close to $80 million from government-related deals.
Early in 2023, Scale laid off about 20% of its workers. Odd timing, since AI demand was heating up, but the wider tech world was in a downturn and Scale trimmed anyway.
Then came the moment that put Scale at the center of the AI boom. It became OpenAI's preferred partner for fine-tuning GPT-3.5, and its services were used to build ChatGPT. Scale also ran red-teaming at DEF CON, testing generative AI for weaknesses, which planted a flag in a new business: evaluating models, not just labeling data.
In 2024 investors piled in. An Accel-led round valued Scale at nearly $13 billion, and soon after it raised another billion from backers including Amazon and Meta, pushing the valuation to $14 billion. At that point Scale was clearly the dominant data-labeling company in the world.
The engine behind all that growth was a crowd of contract workers labeling data, and in late 2024 the cracks showed. A former employee sued, claiming wage theft and that workers had been wrongly classified. A second suit followed, and contractors said tagging disturbing content had hurt them psychologically. The lawsuits dragged the messy human side of the labeling machine into public view.
In mid-2025, Meta agreed to buy a 49% non-voting stake for around $14.8 billion, while Scale stayed independent on paper. Wang left to join Meta, and Jason Droege took over as CEO. Meta was now the biggest shareholder and a guaranteed customer, which made rival AI labs nervous about trusting a company their competitor partly owned.
The nervousness turned out to be justified. Two months after the Meta deal, Scale cut 200 full-time employees and ended work with 500 contractors. Key frontier labs, including OpenAI and Google, paused or shifted work away over the conflict of interest. Scale's private valuation was reported to slide from around $15 billion toward $7–9 billion.
Under CEO Jason Droege, Scale began repositioning, from pure data labeling toward building and deploying AI applications for companies and governments. Labeling still brought in the vast majority of revenue, but Droege expected apps to overtake it within about eighteen months as revenue neared a billion dollars. New customers like BP and Mayo Clinic, plus defense work like the Thunderforge military planning project, anchored the new direction.
To lead that push, Scale turned to Francis deSouza, a former Google Cloud and Illumina executive, as CEO starting in August 2026. He succeeded Droege, who had held the job on an interim basis for about a year since Wang left for Meta. The pick put a seasoned enterprise-tech leader in charge as Scale tried to make reliability and applications its new story.
Scale became one of the most valuable private AI companies, raising money at a reported valuation around $14 billion in 2024, and it was reportedly in talks for a much higher valuation in 2025. It is not publicly traded, so there are no dividends or buybacks to speak of; the money it raises goes into hiring, its software platform, and expanding into new markets. The company has drawn criticism over labor practices in its contractor workforce and over its military work, which some employees have protested. It is a bet on a simple idea: AI is only as good as the data it learns from, and someone has to organize that data.
Saved answer · last researched 4 Nov 2026