Scale AI
Scale AI is a San Francisco-based artificial intelligence company founded in 2016 by Alexandr Wang and Lucy Guo. It supplies data collection, data labeling, human feedback, model evaluation, and applied AI systems to commercial and government customers. The company began as an interface for outsourcing human tasks and later concentrated on the data and evaluation work used to develop machine learning systems.[1][2]
In June 2025, Meta AI made a non-voting minority investment in Scale. Scale said the transaction valued the company at more than $29 billion, while independent reporting described a $14.3 billion investment for a 49% interest.[8][10] Meta's regulatory filing used a different accounting measure: it recorded $13.79 billion of the total consideration as a non-marketable equity investment.[9] The transaction was an investment rather than an acquisition of Scale, and the valuation attached to it is not evidence of a later market value.
History
Scale entered Y Combinator's Summer 2016 program. Its launch announcement described an application programming interface through which customers could send manual tasks such as data extraction and categorization to contractors. Scale said it vetted contractors and performed quality control before returning results.[1] As demand for annotated training data grew, the company focused on datasets for computer vision, autonomous vehicles, and other machine-learning applications.[2]
In December 2020, Scale announced a $155 million Series D financing at a post-money valuation above $3.5 billion. The same announcement introduced Nucleus, a dataset-management and model-testing product, and disclosed the acquisition of Helia AI, a company working on real-time video and AI infrastructure.[2] Scale acquired sensor-data search company SiaSearch in November 2021. The purchase price was not disclosed, and Scale said the SiaSearch team and technology would be incorporated into Nucleus.[3]
In May 2024, Scale closed a $1 billion Series F transaction at a $13.8 billion valuation. The company described the financing as a mixture of primary and secondary transactions.[4] In April 2026, Scale announced the acquisition of ICG Solutions, a defense-technology company specializing in real-time streaming-data analytics, and said ICG would initially operate as a wholly owned subsidiary.[5] These three acquisitions are the acquisitions supported by the cited company announcements and independent reporting; the sources do not establish a complete list of every asset purchase or team hire.
Business and research
Scale's core business combines software with human work to create and manage training and evaluation data. Its services have included labeling images, text, voice, and video, collecting expert-written examples, comparing model outputs, and preparing datasets for model development.[2][3][10] For large language models, human comparisons and corrections can be used in post-training methods such as reinforcement learning from human feedback. The company also develops systems for enterprises and public agencies, including tools for deploying and evaluating models on restricted government networks.[6][16]
Scale launched its Safety, Evaluation, and Alignment Lab in 2023 and expanded the research group as Scale Labs in March 2026. The company described the expanded program as covering model capabilities, agentic and multimodal systems, post-training, deployment, and AI risk evaluation.[6] A peer-reviewed Nature paper published in January 2026 credits the Center for AI Safety, Scale AI, and a contributor consortium with developing Humanity's Last Exam, a 2,500-question, multimodal benchmark of closed-ended academic questions. The paper says the two organizations jointly designed the dataset premise and pipeline and provided funding, infrastructure, and review resources.[7]
Public descriptions of Scale's products and research should not be read as independent validation of performance. Product scope changes over time, and the company does not publish audited counts for many operational claims, including the size of its contributor network, the number of decisions processed, or the revenue attributed to individual product lines.
Meta investment and governance
Scale announced the Meta transaction on June 12, 2025. It said Meta would hold a minority of Scale's outstanding equity, the companies would expand their commercial relationship, and some proceeds would be distributed to shareholders and vested equity holders. Wang joined Meta's AI organization, and Scale said he would remain a director. Scale's board appointed chief strategy officer Jason Droege as interim chief executive officer.[8]
The sources describe the transaction using related but non-identical figures:
| Measure | What the source says |
|---|---|
| Transaction valuation | Scale said the investment valued it at more than $29 billion.[8] |
| Reported investment and ownership | The Associated Press reported $14.3 billion for a 49% interest.[10] |
| Meta's accounting disclosure | Meta reported acquiring a non-voting minority and classified $13.79 billion of the total consideration as a non-marketable equity investment under the measurement alternative.[9] |
Meta's filing said it used that accounting treatment because it did not have significant influence over Scale's operations.[9] That is an accounting conclusion, not a statement that no commercial or governance conflicts could arise. Scale separately said it remained independent, would not give Meta access to internal systems or customer-confidential information, and would use recusal and conflict-of-interest procedures for Wang's board role.[11] Those safeguards are Scale's representations.
The investment affected customer perceptions. Reuters reported in June 2025, citing five people familiar with the matter, that Google planned to end its Scale relationship and was discussing work with competitors after the Meta deal.[12] This was a reported plan, not proof that every proposed transition was completed. In July 2025, Scale cut about 200 employees, approximately 14% of staff, and ended engagements with about 500 contractors. Droege's internal memo attributed the restructuring to expanding the core data-labeling business too quickly and said the company would add staff in enterprise and government units.[13]
Government and defense work
Scale has received United States defense work involving annotated datasets, model evaluation, and decision-support software. Contract ceiling, award value, and money obligated are different measures and should not be treated as interchangeable.
| Date | Record | Financial scope |
|---|---|---|
| September 2020 | The Department of Defense awarded Scale a firm-fixed-price research contract to develop methods for producing high-quality annotated AI datasets.[14] | The contract's face value was $90,865,236; $16,472,937 was obligated at award.[14] |
| March 2025 | The Defense Innovation Unit awarded Scale a prototype contract for Thunderforge, a system intended to support operational and theater-level military planning. The announcement also named Anduril and Microsoft as technology participants.[15] | The public announcement did not state a dollar value.[15] |
| September 2025 | The Chief Digital and Artificial Intelligence Office entered a five-year production agreement for access to Scale's data, evaluation, and deployment platforms.[16] | Axios reported a $100 million ceiling and a $40.7 million initial project commitment.[16] |
| May 2026 | The Defense Department increased the production agreement's ceiling after additional components began using the vehicle.[17] | The ceiling rose from $100 million to $500 million. A ceiling is a maximum potential value, not an amount already paid or obligated.[17] |
These records support a substantial defense business but not the claim that every ceiling amount became revenue for Scale. They also do not support describing Thunderforge as an autonomous weapons system; the Defense Innovation Unit described it as planning and decision-support infrastructure.[15]
Contractor workforce and labor disputes
Human contributors perform much of the annotation and evaluation work behind Scale's services. A 2023 Washington Post investigation of Remotasks workers in the Philippines reported delayed, reduced, or cancelled payments and limited avenues for recourse. Scale said its payment system was improving and that delays or interruptions were extremely rare.[18] The report documented disputed working conditions; it did not establish a verified global headcount for Scale's contributors.
Fairwork's 2025 Cloudwork Ratings gave Remotasks 2 points out of 10, awarding points for measures intended to mitigate overwork and health and safety risks. Fairwork said a missing point means that its researchers did not find enough evidence to verify a threshold, not necessarily that the platform violated it. The report also documented changes by Remotasks, including base pay for task-loading time, secondary assignments, advance notice of contract changes, and an appeal channel for account deactivation. Scale disputed the score as overly subjective while acknowledging some of the changes.[19]
Several workers brought wage-and-hour claims against Scale in 2024 and 2025, including allegations that contributors were misclassified as independent contractors and were not paid for all working time. Scale denied the allegations and said its practices complied with applicable law.[20] The United States Department of Labor separately investigated Scale's compliance with the Fair Labor Standards Act and closed the investigation in May 2025 without publicly stating a reason.[21]
As of July 28, 2026, a California court had preliminarily approved a proposed settlement in McKinney et al. v. Scale AI, Inc., et al. The court had not granted final approval, and a hearing was scheduled for October 30, 2026. The defendants denied violating wage laws and denied that contributors were employees rather than independent contractors.[22] Preliminary settlement approval is not a judicial finding that the allegations were proven.
See also
- Data labeling
- Model Evaluation
- Reinforcement Learning from Human Feedback
- Alexandr Wang
- Humanity's Last Exam
References
- ^Scale is an API for human labor - Y Combinator, July 26, 2016.
- ^Scale AI's Series D: Expanding our Team to Empower AI Dreamers - Scale AI, December 1, 2020.
- ^Scale AI acquires SiaSearch to give companies the same data engine superpowers that Tesla has - TechCrunch, November 3, 2021.
- ^Scale's Series F: Expanding the Data Foundry for AI - Scale AI, May 21, 2024.
- ^Scale AI Acquires ICG Solutions to Accelerate National Security AI - Scale AI, April 20, 2026.
- ^Introducing Scale Labs - Scale AI, March 9, 2026.
- ^A benchmark of expert-level academic questions to assess AI capabilities - *Nature*, published online January 28, 2026.
- ^Scale AI Announces Next Phase of Company's Evolution - Scale AI, June 12, 2025.
- ^Meta Platforms, Inc., Quarterly Report for the quarter ended June 30, 2025 - U.S. Securities and Exchange Commission.
- ^Meta invests $14.3B in AI firm Scale and recruits its CEO for 'superintelligence' team - Associated Press, June 13, 2025.
- ^What Meta's Investment Means for Our Customers, Partners, and Contributors - Scale AI, June 18, 2025.
- ^Google, Scale AI's largest customer, plans split after Meta deal, sources say - Reuters, June 13, 2025.
- ^Scale AI lays off 14% of staff, largely in data-labeling business - TechCrunch, July 16, 2025.
- ^Contracts for September 29, 2020 - U.S. Department of Defense, September 29, 2020.
- ^DIU's Thunderforge Project to Integrate Commercial AI-Powered Decision-Making for Operational and Theater-Level Planning - Defense Innovation Unit, March 5, 2025.
- ^Exclusive: Scale AI strikes deal with Pentagon - Axios, September 17, 2025.
- ^DOD grows Scale AI agreement to $500M - Washington Technology, May 7, 2026.
- ^Behind the AI boom, an army of overseas workers in 'digital sweatshops' - *The Washington Post*, August 28, 2023.
- ^Fairwork Cloudwork Ratings 2025 - Fairwork, University of Oxford and WZB Berlin Social Science Center, May 2025.
- ^Scale AI hit by its second employee wage lawsuit in less than a month - TechCrunch, January 9, 2025.
- ^US Department of Labor drops investigation into Scale AI - Reuters, May 9, 2025.
- ^McKinney, et al. v. Scale AI, Inc., et al. settlement website - Court-authorized settlement administrator, 2026.
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.
6 revisions · v7 · 1,861 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Independent 2026-07-28 fact-check: 22 official, primary, peer-reviewed, academic-labor, and independent sources and 40 citation calls checked; root review separately rechecked the Meta transaction measures, HLE and Scale Labs scope, ICG acquisition, defense agreement ceilings and obligations, Fairwork score and caveat, preliminary settlement posture, cutoff-bounded leadership, all 11 direct internal targets, and all 22 article/source renders.
Cite this page: AI Wiki. "Scale AI." aiwiki.ai, updated 30 Jul 2026, fact-checked 30 Jul 2026. CC BY 4.0. https://aiwiki.ai/wiki/scale_ai