Scale AI
| Company type | Private |
|---|---|
| Industry | Artificial intelligence, Data annotation, Software development |
| Founded | 2016 |
| Founder | Alexandr Wang, Lucy Guo |
| Headquarters | San Francisco , United States |
Area served | Worldwide |
Key people | Jason Droege (CEO), Alexandr Wang (senior leadership at Meta Platforms), Michael Kratsios (managing director and head of strategy) |
| Products | Data annotation services, LLM evaluation, AI software development, Remotasks, Outlier |
| Parent | None |
Scale AI, Inc. is an American company headquartered in San Francisco that specializes in data annotation services, including data labeling, evaluation of large language models (LLMs), and software development for AI applications.
Its research arm, known as the Safety, Evaluation and Alignment Lab, is dedicated to the assessment and alignment of LLMs through initiatives such as Humanity's Last Exam, which serves as a benchmark for AI safety, reasoning, and alignment. Scale AI conducts data labeling operations through its subsidiaries: Remotasks, which focuses on computer vision and autonomous vehicle projects, and Outlier, which concentrates on annotating data for LLMs.
The company’s client portfolio includes leading technology and automotive corporations like Google, Microsoft, Meta, General Motors, OpenAI, and Time. Additionally, Scale AI collaborates with governmental bodies, including the U.S. government on defense-related initiatives and Qatar to improve the effectiveness of social programs.
Founding and Early Development (2016–2019)
Established in 2016 by Alexandr Wang and Lucy Guo through Y Combinator, Scale AI was born from the founders’ prior experience working together at Quora. Initial funding came from investors such as Dragoneer Investment Group, Tiger Global Management, and Index Ventures. Lucy Guo departed from the company in 2018.
In August 2019, following a $100 million investment from Peter Thiel’s Founders Fund, Scale achieved a valuation exceeding $1 billion, officially becoming a unicorn startup.
Expansion Phase (2019–2025)
In 2020, Scale secured a contract with the United States Department of Defense.
By May 2021, former U.S. Chief Technology Officer under the Trump administration, Michael Kratsios, joined Scale AI as managing director and head of strategy.
Scale’s valuation surged to $7 billion by July 2021 after a funding round led by Greenoaks, Dragoneer Investment Group, and Tiger Global Management, reflecting increasing demand for data annotation across various sectors.
In January 2022, Scale was awarded a $250 million contract to provide federal agencies in the United States with access to its AI technologies. The following month, it introduced an Automated Damage Identification Service that analyzes satellite images to evaluate structural damage in Ukraine amid the Russian invasion. This innovation was recognized by Time magazine as one of the Best Inventions of 2022. That year also saw Scale expand its presence by opening an office in St. Louis.
January 2023 brought a workforce reduction of 20%. By May, Scale partnered with the US Army’s XVIII Airborne Corps, becoming the first AI company to deploy its large language model, Donovan, within a classified military network.
In August 2023, Scale entered into a partnership with OpenAI as the preferred collaborator for fine-tuning GPT-3.5. The company’s contributions also supported the early development stages of ChatGPT. That same month, Scale’s evaluation platform was utilized during DEF CON’s inaugural generative AI red team event.
By December 2023, Scale joined Meta Platforms’s Purple Llama initiative aimed at advancing secure open generative AI development.
In February 2024, the Department of Defense selected Scale for a one-year contract focused on testing and evaluating LLMs intended for military applications.
March 2024 marked Scale’s valuation approaching $13 billion after securing funding led by Accel. By May 2024, an additional $1 billion investment from entities including Amazon and Meta Platforms elevated its valuation to $14 billion.
In August 2024, Scale formalized a collaboration agreement with the US AI Safety Institute to jointly conduct research and evaluation of AI models. This institute operates under the auspices of the Department of Commerce’s National Institute of Standards and Technology.
Late 2024 into early 2025 saw legal challenges arise with multiple lawsuits accusing Scale of wage theft and causing psychological distress related to workplace conditions.
Early in 2025, reports revealed cooperation between Scale and Meta on Defense Llama, a military-oriented LLM product. During this period, Scale ran a full-page advertisement urging President Donald Trump to "win the AI war." The company also partnered with the Center for AI Safety to launch Humanity's Last Exam, an AI benchmarking test. Additionally, it contributed to other benchmark projects such as EnigmaEval, MultiChallenge, and MASK.
In February 2025, Scale signed a five-year agreement with Qatar’s government to enhance public service delivery through AI-driven tools like predictive analytics and automation. This contract was formalized at the Web Qatar 2025 Summit by Qatar’s Minister of Communications and Information Technology. That same month, Scale became an official third-party evaluator for the U.S. AI Safety Institute.
March 2025 brought another Department of Defense contract for the Thunderforge initiative aimed at leveraging AI to efficiently plan and execute military asset deployments. This project includes partners such as Anduril Industries and Microsoft, targeting operations within USINDOPACOM and EUCOM commands.
In April 2025, Scale introduced Scale Evaluation—a platform designed to benchmark LLMs by detecting vulnerabilities and identifying opportunities for improving training datasets.
Meta's Additional Investment (2025–present)
On June 10, 2025, Meta Platforms committed to acquiring a 49% ownership interest in Scale AI for $14.8 billion. This strategic move was intended to secure access to specialized datasets that would bolster the development of Meta’s Llama series of large language models (LLMs). Despite this transaction, Scale AI continues to operate as an independent company. Former CEO Alexandr Wang transitioned to a senior leadership position within Meta, while Jason Droege—previously Scale AI’s chief strategy officer and an ex-executive at Uber—was appointed as the new CEO. In response to the acquisition, Google, which had been Scale AI’s largest client, declared its intention to discontinue the partnership.
Remotasks
Established in 2017, Remotasks serves as Scale AI’s crowdworking platform dedicated to producing annotated datasets for machine learning purposes, including applications in computer vision and autonomous driving technologies. The platform maintains operational centers across Southeast Asia and Africa. In 2019, Scale AI launched Smart Ecosystem Philippines to oversee Remotasks’ local activities.
The platform predominantly depends on freelance workers who frequently lack formal labor protections. Expansion into markets such as India and Venezuela intensified competition, driving compensation for annotation tasks below one cent per task in some cases. Investigations have revealed recurrent delays in payments and instances where workers received only a portion of their agreed earnings. A 2022 study by the Oxford Internet Institute evaluated Remotasks against fair work criteria and determined it satisfied just one out of ten standards.
Criticism has been directed at Remotasks for obscuring its affiliation with Scale AI, poor communication practices, and abrupt restrictions on worker access. Administrative challenges led to the suspension of Remotasks operations in early 2024 across several countries, including Kenya, Nigeria, and Pakistan.
Outlier
Outlier is a contributor-driven platform managed by Scale AI that specializes in data tasks related to generative AI, particularly those involved in building and refining large language models (LLMs). Contributors are generally highly qualified professionals holding advanced degrees, possessing industry experience, and exhibiting native-level fluency across multiple languages. The platform’s activities encompass content assessment and reinforcement learning from human feedback (RLHF), playing a vital role in training AI models for leading corporations.