Scale AI
AI training data and evaluation for enterprise and government
- Overview
- Scale AI provides data labeling, model evaluation, and post-training infrastructure for AI development, with a growing focus on government contracts and physical AI. Since Meta acquired a 49% non-voting stake in June 2025 and founder Alexandr Wang moved to Meta, frontier-lab customers including Google, OpenAI, and xAI have cut or reduced contracts; Scale has pivoted toward U.S. defense and robotics. Francis deSouza was named CEO in August 2026, and the company filed an S-1 with the SEC in September 2026.
- Why it matters
- Scale AI was the default data infrastructure vendor for frontier AI labs, but Meta's $14.3B investment in June 2025 ended that neutrality: Google, OpenAI, and xAI all pulled or sharply reduced contracts, shifting business to competitors like Mercor and Surge AI. The company has reoriented toward U.S. defense — landing a $500M Pentagon contract in May 2026 — and physical AI for robotics, which now define its growth thesis. Practitioners evaluating Scale should weigh the conflict-of-interest risk if their own training data would flow through a Meta-affiliated vendor, and monitor the pending IPO for financial transparency.
- In practice
- Enterprise AI teams use Scale's Data Engine to generate labeled datasets for fine-tuning and RLHF, and its GenAI Platform for evaluation and red-teaming before model releases. Government and defense programs use Scale for classified and sensitive data annotation workflows where cleared annotators are required. The honest limitation is that any AI lab competing with Meta should assume a structural conflict of interest and consider vendor diversification; Scale's SEAL benchmarks remain well-regarded and are widely cited independently of the ownership question.
- Best for
- Government agencies and enterprises building or evaluating AI systems that need high-quality labeled data, model evaluation, or physical/robotics AI training pipelines — particularly those without a conflict-of-interest concern around Meta's ownership stake.
- Pricing
- Self-serve tier (Scale Data Engine/Rapid): free allocation of 1,000 labeling units + 10,000 images, then pay-as-you-go at $0.05/unit for self-labeling. Enterprise Data Engine and GenAI Platform are custom-priced with no public rate card; enterprise contracts reportedly average ~$93,000/year and can exceed $400,000 for complex projects.
Key features
- Human-in-the-loop and AI-assisted data labeling across image, video, text, audio, and sensor modalities
- RLHF and post-training data pipelines for LLM fine-tuning
- Model evaluation and red teaming via the SEAL benchmarks (Humanity's Last Exam, MultiChallenge, MASK)
- Physical AI Data Engine for robotics and industrial deployment
- RL Environments for training and evaluating agents in simulated tool-use and computer-use workflows
Quick answers
- What is Scale AI?
- Scale AI provides data labeling, model evaluation, and post-training infrastructure for AI development, with a growing focus on government contracts and physical AI. Since Meta acquired a 49% non-voting stake in June 2025 and founder Alexandr Wang moved to Meta, frontier-lab customers including Google, OpenAI, and xAI have cut or reduced contracts; Scale has pivoted toward U.S. defense and robotics. Francis deSouza was named CEO in August 2026, and the company filed an S-1 with the SEC in September 2026.
- Why does Scale AI matter?
- Scale AI was the default data infrastructure vendor for frontier AI labs, but Meta's $14.3B investment in June 2025 ended that neutrality: Google, OpenAI, and xAI all pulled or sharply reduced contracts, shifting business to competitors like Mercor and Surge AI. The company has reoriented toward U.S. defense — landing a $500M Pentagon contract in May 2026 — and physical AI for robotics, which now define its growth thesis. Practitioners evaluating Scale should weigh the conflict-of-interest risk if their own training data would flow through a Meta-affiliated vendor, and monitor the pending IPO for financial transparency.
- What is Scale AI best for?
- Government agencies and enterprises building or evaluating AI systems that need high-quality labeled data, model evaluation, or physical/robotics AI training pipelines — particularly those without a conflict-of-interest concern around Meta's ownership stake.
- How much does Scale AI cost?
- Self-serve tier (Scale Data Engine/Rapid): free allocation of 1,000 labeling units + 10,000 images, then pay-as-you-go at $0.05/unit for self-labeling. Enterprise Data Engine and GenAI Platform are custom-priced with no public rate card; enterprise contracts reportedly average ~$93,000/year and can exceed $400,000 for complex projects.
Seen in recent stories
Where Scale AI showed up in the last 90 days of KeyNews editions.
Know the terms
Glossary entries that explain what Scale AI is built on.
Reinforcement Learning from Human Feedback (RLHF)
A training technique where human raters rank model outputs, and the model learns to prefer higher-ranked responses. RLHF is what makes AI assistants helpful, harmless, and conversational rather than just autocomplete.
Evals
Systematic evaluation frameworks that measure AI model performance on specific tasks relevant to your use case, going beyond generic benchmarks to test the behaviors that actually matter for your application.
Red teaming
The practice of systematically probing an AI system to find vulnerabilities, biases, and failure modes before deployment. Red teaming is now standard practice at major AI labs and increasingly required by regulation.
Synthetic data
Artificially generated training data created by AI models or simulations. Synthetic data is increasingly used when real data is scarce, private, or expensive, but quality and diversity remain open challenges.
More enterprise ai tools
New tools drop every week.
ONE BRIEFING · EVERY FRIDAY · FREE
Free. Unsubscribe anytime.