Scale AI

Scale AI supplies expert training data, RLHF and model evaluation to AI developers through its Data Engine, and sells agent-building platforms to enterprises (GenAI Platform) and governments (Donovan). It suits organisations that need high-quality labelled data or governed AI deployments in sensitive environments, including classified defence networks.

Scale AI, founded in 2016 by Alexandr Wang and Lucy Guo and based in San Francisco, started as a data-labelling company and now operates across three lines. The Data Engine provides data collection, curation and annotation across text, image, video and 3D sensor fusion, plus generative-AI services such as prompt-response pair creation, RLHF, red teaming and model evaluation; Scale says it supplies training data to 90% of leading generative-AI model developers and publishes the SEAL leaderboards. The GenAI Platform lets enterprises connect data sources such as Confluence, SharePoint and S3, build and orchestrate long-running multi-agent workflows on models from OpenAI, Google, Meta, Mistral and others, evaluate and monitor them, and feed human feedback back in, deployed inside the customer's own VPC on AWS, Azure or GCP. Donovan is its no-code agent platform for the public sector, authorised at DISA IL4 and FedRAMP High, field-ready on classified networks, and used by customers including the US Army, US Air Force and CDAO. In June 2025 Meta paid $14.3 billion for a 49% non-voting stake and Wang left to lead Meta's superintelligence unit; OpenAI and Google reportedly wound down work with Scale afterwards. Jason Droege served as interim CEO until Francis deSouza, formerly Google Cloud COO, took over on August 10, 2026. In May 2026 Scale expanded its Pentagon CDAO agreement from $100 million to $500 million, and it cites enterprise wins including BP and Mayo Clinic.

Category

Data & Analytics

Pricing

Contact for pricing

Target Market

ML Engineering Leaders, CTOs, Government and Defence Agencies, Enterprise AI Leaders

Deployment

Multi-cloud, Hybrid

Founded

2016

Headquarters

San Francisco, California, USA

Team Size

500+

• ✓
  Data Engine

  Collects, curates and annotates text, image, video and 3D sensor-fusion data with expert labellers for model training at scale.

• ✓
  RLHF and red teaming

  Generates prompt-response pairs, human preference data and adversarial tests used to align and harden generative models before release.

• ✓
  GenAI Platform

  Builds, orchestrates, evaluates and monitors enterprise agents over connected data, deployed in the customer's own VPC on AWS, Azure or GCP.

• ✓
  Scale Donovan

  No-code public-sector agent platform with an agent factory, model comparison and pre-built intelligence agents, authorised at IL4 and FedRAMP High.

• ✓
  SEAL leaderboards and evaluation

  Publishes independent model capability benchmarks and runs custom evaluations, helping buyers compare models on accuracy and safety.

• ✓
  Model flexibility

  Supports models from OpenAI, Google, Meta, Mistral and others with switching, so enterprises avoid lock-in to one provider.

• •
  Frontier model post-training

  A model developer buys expert-written prompt-response pairs and preference rankings to fine-tune and align a new generative model before launch.

• •
  Defence decision support

  Department of War components use Donovan and the GenAI Platform under a $500 million CDAO agreement for decision-support and data operations.

• •
  Perception data labelling

  An autonomous-systems team sends camera and lidar data through the Data Engine to get labelled 3D sensor-fusion training sets.

• •
  Enterprise agent deployment

  An energy company like BP works with Scale to build agents over internal documents that run in its own cloud VPC with monitoring.

• •
  Model evaluation before procurement

  An enterprise compares candidate models on accuracy, hallucination rate and latency before choosing which to deploy in production.

Scale publishes no list prices; its pricing page only offers a 1:1 intro call. Data work is scoped per project, platform deals are negotiated, and US government buyers can contract through the CDAO Production OTA, which lets any Department of War component start a project agreement without a new competitive solicitation.

This page was written from 9 sources, 3 on domains other than scale.com.

1. 1.scale.com — scale.comvendor
2. 2.scale.com — genai platformvendor
3. 3.scale.com — donovanvendor
4. 4.scale.com — data enginevendor
5. 5.scale.com — scale appoints new ceovendor
6. 6.scale.com — Scale ai pentagon cdao 500 million agreementvendor
7. 7.time.com — meta scale ai data industry
8. 8.en.wikipedia.org — Scale AI
9. 9.hn.algolia.com — hn.algolia.com