Summary

Scale AI is a data infrastructure company focused on providing high-quality training data, data labeling, model evaluation, and AI safety tooling for machine learning teams. Founded in 2016 by Alexandr Wang, Scale has grown into a leading provider of annotation services, managed labeling platforms (Remotasks, Outlier), and model evaluation products used by major tech companies and government agencies.

Key facts

  • Founded: 2016 (Y Combinator Summer 2016)
  • Headquarters: San Francisco, CA
  • Founders: Alexandr Wang (CEO until June 2025)
  • Interim CEO: Jason Droege (appointed June 2025)
  • Employees: ~900 (2025)
  • Gig workforce: 240,000+ annotators (Remotasks/contractors)
  • 2024 revenue: 2B
  • Total funding pre-2025: ~$1.6B

Scale provides:

  • Human-in-the-loop data labeling and annotation (images, video, text)
  • Synthetic data generation and tools for LLM training (Outlier)
  • Model evaluation, safety, and alignment benchmarks (SWE-Bench, Humanity’s Last Exam)
  • Enterprise AI solutions: custom models and agents using customer data
  • Managed workforce and platform integrations for large-scale labeling programs

Recent developments (2024-2025)

  • May 2024 valuation: $13.8B after funding rounds.
  • June 12, 2025: Meta Platforms announced a major investment valuing Scale at over 14B).
  • Leadership change: Jason Droege named Interim CEO after Wang’s departure to Meta.
  • July 2025: Scale announced layoffs (~200 employees) from its labeling operations, citing shifts in market demand; simultaneously won a $99M contract with the U.S. Army.
  • August 2025 onward: Reports of operational friction with Meta and TBD Labs; some internal clients at Meta explored alternative suppliers, raising questions about integration and data-quality expectations.

Customers & partners

Notable customers and partners include Google, Microsoft, Meta, OpenAI, General Motors, Toyota Research Institute, and several U.S. government agencies (DoD, U.S. Army).

Why it matters

Scale sits at a strategic point in the AI stack: high-quality labeled data is a critical bottleneck for training and evaluating high-performance AI models. Scale’s combination of tooling, workforce, and enterprise services makes it a key supplier for companies building foundation models and specialized AI systems.

Risks & Critiques

  • Quality vs. scale tradeoffs: crowdsourced labeling can struggle to meet the specialized needs of frontier AI, which increasingly demands expert-labeled data and domain knowledge.
  • Competitive pressure: other data-labeling firms (Labelbox, Appen alternatives, Mercor, Surge) and in-house labeling teams at large AI labs compete for customers.
  • Strategic concentration: Meta’s large minority stake raised concerns about data access and competitive neutrality, leading some customers (notably Google) to reconsider commercial ties.

Sources & further reading

  • Scale AI official announcements and blog (June 2025)
  • Press coverage: TechCrunch, Bloomberg, Reuters, NYTimes (2024-2025)
  • Industry analysis and podcast coverage (a16z, Y Combinator, various AI news shows)