Key facts

Founded
2016
HQ
San Francisco, United States
Employees
1,000+
Website
Funding stage
Strategic
Further facts
  • Capital: Scale raised US$1bn in a May 2024 Series F led by Accel at a US$13.8bn valuation. In June 2025 Meta made a strategic investment of an undisclosed amount, valuing Scale at over US$29bn; Scale is using the proceeds for product work and customer partnerships and will distribute proceeds to shareholders and vested equity holders.
  • Ownership: Meta holds a minority, non-voting stake of about 49% alongside an expanded commercial relationship. Scale runs as an independent company with no integration of operations with Meta and no Meta access to customer data.
  • Timeline: Founded in 2016 by Alexandr Wang, who joined Meta in June 2025 and remains Scale's chairman. Jason Droege led the company as interim CEO from June 2025, and former Google Cloud COO and Illumina CEO Francis deSouza became CEO on 10 August 2026. In October 2026 Scale set up a long-term base at The Pearl in Doha and launched a UAE entity at the Dubai International Financial Centre.
  • Offices: Headquarters in San Francisco, with offices in New York, Washington DC, St. Louis, London, Doha and Dubai and a presence in Riyadh.
  • Scale: Scale has generated more than 15 billion human decisions for AI training and paid more than US$1bn to its contributors. It signed well over US$1bn in new business in 2025, holds two US Department of War contracts worth nearly US$200m combined, and has trained or upskilled nearly 6,000 people in Qatar.

Focus areas

Reinforcement learning Data labelling LLM evaluation AI agents Foundation models Defence Physical AI Sovereign AI

About

Scale AI provides the human-generated data, evaluations and software that AI labs, enterprises and governments use to build and run AI systems. Frontier model builders use its Data Engine for RLHF, red-teaming and evaluation data produced by expert contributors, a quarter of whom have advanced degrees. Enterprises and public bodies use its GenAI Portfolio (SGP) and Donovan to build, test and deploy agents on their own data, with customers including Mayo Clinic, BP, TIME and the Pentagon's Chief Digital and AI Office. Its research arm publishes benchmarks such as Humanity's Last Exam and SWE-Bench Pro. Founded in 2016 and based in San Francisco, Scale has more than 1,000 employees and offices across the US, London and the Gulf. Meta took a minority, non-voting stake in June 2025 that valued the company at more than US$29bn, and Francis deSouza became chief executive in August 2026.

How they use AI

The Data Engine pairs model-assisted labelling with expert human contributors to produce prompt-response pairs, RLHF preference data, red-teaming attacks and evaluation sets across text, documents, images, video and 3D sensor data. SGP connects to enterprise sources such as SharePoint, Confluence and S3, deploys agents inside a customer's own AWS, Azure or GCP environment, works across OpenAI, Google, Meta and Mistral models, and feeds human feedback and production telemetry back in as a training signal. Every production agent carries an audit trail and source-cited outputs. Scale Post-Training fine-tunes open-weight models such as NVIDIA Nemotron for specific enterprise jobs to cut inference cost. Donovan applies the same tooling to defence and intelligence work, with a no-code Agent Factory, side-by-side model testing on accuracy and hallucination rates, and a catalogue of mission agents.

Products and services

  • Scale Data Engine — Expert-labelled training, RLHF, red-teaming and evaluation data for frontier and generative models
  • Scale GenAI Portfolio (SGP) — Platform to connect enterprise data and build, deploy, evaluate and improve agents in a customer's own cloud
  • Scale Post-Training — Customises open-weight models such as NVIDIA Nemotron into task specialists
  • Scale Donovan — Agent platform for defence and national-security teams, including classified data workflows
  • Scale Labs evaluations — Benchmarks and leaderboards such as Humanity's Last Exam, SWE-Bench Pro and MCP Atlas
  • Physical AI data — Real-world robotics and autonomy data for training robotic foundation models

Notable customers / partners

  • Chief Digital and AI Office (US Department of War)
  • Mayo Clinic
  • BP
  • Allianz
  • TIME
  • Howard Hughes
  • Physical Intelligence
  • Universal Robots
  • Cengage
  • Shore Capital
  • EY
  • Skydance
  • OpenAI
  • Microsoft
  • Qatar Ministry of Communications and Information Technology

Key people

  • Francis deSouza — Chief Executive Officer
  • Alexandr Wang — Founder and Chairman of the Board

Investors

  • Accel
  • Meta
  • Y Combinator
  • Index Ventures
  • Founders Fund
  • Coatue
  • Thrive Capital
  • Spark Capital
  • NVIDIA
  • Tiger Global
  • Greenoaks
  • Wellington Management
  • Cisco Investments
  • DFJ Growth
  • Intel Capital
  • ServiceNow Ventures
  • AMD Ventures
  • WCM
  • Nat Friedman