Scale AI — research brief

Slug: scale-ai
Industry: AI Data / Evaluation
Last updated: 2026-09-20 (America/Chicago)
Teaching arc: Data labeling → GenAI platform / evaluation → Enterprise / Government (Donovan) → partnerships (Meta, CSPs)

Label key: FACT = primary (Scale site, docs, company PR, AWS Marketplace listing). SECONDARY = reputable press. INFERENCE = model reading from facts (not a cited metric).

Company snapshot

Scale AI (scale.com) sells AI data infrastructure and enterprise GenAI tooling: human + automated annotation, generative-AI data engines (RLHF / preference data / expert data), evaluation & red-teaming (SEAL), the Scale GenAI Platform (SGP) for agent build/deploy/monitor, and Donovan for public-sector AI agents. Founded 2016; homepage claims the models at the frontier run on Scale data and that “90% of the world's leading generative AI model builders are powered by Scale” (FACT — scale.com homepage marketing claim).

Products (FACT — scale.com)

Product What it is Primary buyers
Scale Data Engine / Generative AI Data Engine Expert-curated training & preference data; Ops Center QC; RLHF / ranking Frontier labs, ML teams
Self-Serve Data Engine Annotate + manage; BYO workforce; card PAYG Research / experimental teams
Scale Studio Bring-your-own labelers on Scale’s platform Teams with in-house annotators
Rapid (docs) Self-serve labeling API/UI historically documented Prototyping / small batches
Scale GenAI Platform (SGP) Connect enterprise data → build/execute agents → evaluate/monitor → learn Enterprises deploying GenAI agents
SEAL / Leaderboards / Showdown Safety, Evaluations, and Alignment Lab; private & preference leaderboards Labs + public signal
Donovan Public-sector AI agents; classified / air-gapped; FedRAMP High / DISA IL4 Defense, intel, federal civilian

Pricing (prefer official)

Official pricing page (FACT — https://scale.com/pricing)

Rapid self-labeling docs (FACT — Scale API Rapid FAQ)

Note: Prefer the public pricing page free-unit figures (1,000 / 10,000) for marketing summary; cite Rapid FAQ when discussing the documented $0.05 self-label unit.

AWS Marketplace list prices (FACT — AWS listings; list, not negotiated)

Meta partnership & valuation

Homepage also states “Partnering to accelerate Meta's LLM and Generative AI” (FACT — marketing).

Government / compliance (FACT — Donovan / SGP pages)

Evaluation brand (FACT — Scale blogs)

What we do not invent

Teaching arc (INFERENCE from product ladder)

  1. Land: Self-serve / Rapid labeling + free units → experiment with annotation & data management.
  2. Expand data: Managed Data Engine / Generative AI Data Engine + expert workforce for frontier training data.
  3. Evaluate: SEAL / Evaluation / red-teaming as trust layer for labs and enterprises.
  4. Platform: Scale GenAI Platform — agents over enterprise data with eval flywheel.
  5. Gov & partners: Donovan for public sector; CSP marketplaces (AWS); strategic Meta commercial expansion.

Primary sources

# Scale AI — research brief

**Slug:** `scale-ai`  
**Industry:** AI Data / Evaluation  
**Last updated:** 2026-09-20 (America/Chicago)  
**Teaching arc:** Data labeling → GenAI platform / evaluation → Enterprise / Government (Donovan) → partnerships (Meta, CSPs)

Label key: **FACT** = primary (Scale site, docs, company PR, AWS Marketplace listing). **SECONDARY** = reputable press. **INFERENCE** = model reading from facts (not a cited metric).

## Company snapshot

Scale AI (scale.com) sells AI data infrastructure and enterprise GenAI tooling: human + automated annotation, generative-AI data engines (RLHF / preference data / expert data), evaluation & red-teaming (SEAL), the Scale GenAI Platform (SGP) for agent build/deploy/monitor, and Donovan for public-sector AI agents. Founded 2016; homepage claims the models at the frontier run on Scale data and that “90% of the world's leading generative AI model builders are powered by Scale” (**FACT** — scale.com homepage marketing claim).

## Products (FACT — scale.com)

| Product | What it is | Primary buyers |
|--------|------------|----------------|
| Scale Data Engine / Generative AI Data Engine | Expert-curated training & preference data; Ops Center QC; RLHF / ranking | Frontier labs, ML teams |
| Self-Serve Data Engine | Annotate + manage; BYO workforce; card PAYG | Research / experimental teams |
| Scale Studio | Bring-your-own labelers on Scale’s platform | Teams with in-house annotators |
| Rapid (docs) | Self-serve labeling API/UI historically documented | Prototyping / small batches |
| Scale GenAI Platform (SGP) | Connect enterprise data → build/execute agents → evaluate/monitor → learn | Enterprises deploying GenAI agents |
| SEAL / Leaderboards / Showdown | Safety, Evaluations, and Alignment Lab; private & preference leaderboards | Labs + public signal |
| Donovan | Public-sector AI agents; classified / air-gapped; FedRAMP High / DISA IL4 | Defense, intel, federal civilian |

## Pricing (prefer official)

### Official pricing page (FACT — https://scale.com/pricing)

- **Enterprise:** sales-led; enterprise-grade quality & SLAs; access to **both** Data Engine and Enterprise GenAI Platform; dedicated customer ops. Data Annotation by customer workforce **or** Scale’s + Data Management; GenAI Portfolio for customized enterprise GenAI apps.
- **Self-Serve Data Engine:** pay as you go via credit card.
  - Data Annotation (BYO workforce): **first 1,000 labeling units at no cost**
  - Data Management: **first 10,000 images at no cost**
- List prices for managed annotation, expert GenAI data, and SGP are **not** published on the public pricing page (contact sales).

### Rapid self-labeling docs (FACT — Scale API Rapid FAQ)

- 200 free labeling units / month for Rapid self-labeling; **$0.05 per additional labeling unit** (1 annotation = 1 unit; 1 attribute = 0.5 units).
- Studio = bring your own annotators on the platform.

Note: Prefer the public **pricing page** free-unit figures (1,000 / 10,000) for marketing summary; cite Rapid FAQ when discussing the documented $0.05 self-label unit.

### AWS Marketplace list prices (FACT — AWS listings; list, not negotiated)

- **Scale Donovan:** contract license **$1,000,000 / month** for up to **250 users** per account (AWS Marketplace).
- **Scale GenAI Platform (SaaS):** 1-month contract dimensions include **Platform Fee $1,000,000 / month** and **Scale Token Units $1,000,000 / month**; longer contracts advertise discounts up to ~92–97% (AWS Marketplace). Treat as **list / marketplace** figures — enterprise deals may differ.

## Meta partnership & valuation

- **FACT (Scale company PR, Jun 12, 2025):** Meta invested; deal **values Scale at over $29 billion**; expands commercial relationship for Scale data solutions; founder Alexandr Wang joining Meta AI efforts while remaining a Scale director; Jason Droege appointed Interim CEO; Meta holds a **minority** of outstanding equity; Scale remains independent. Source: https://scale.com/blog/scale-ai-announces-next-phase-of-company-evolution
- **SECONDARY (Reuters / CNBC / AP):** Meta stake framed as ~**$14.3B for ~49% non-voting**; Wang to Meta superintelligence team. Use only as secondary color; prefer Scale PR for valuation floor.

Homepage also states “Partnering to accelerate Meta's LLM and Generative AI” (**FACT** — marketing).

## Government / compliance (FACT — Donovan / SGP pages)

- Donovan: unclassified/CUI GovCloud with **DISA IL4** and **FedRAMP High Authorized**; classified / air-gapped deployments; model-agnostic; Agent Factory / evaluations / Defense Llama mentioned.
- SGP security claims: DoD IL4 Provisional Authorization, SOC 2 Type II, ISO 27001, FedRAMP High Authorized.

## Evaluation brand (FACT — Scale blogs)

- SEAL = Safety, Evaluations, and Alignment Lab; private leaderboards that cannot be gamed; Showdown preference leaderboard; evaluation partner narrative including NIST / DoD / DEF CON red-teaming (blog claims).

## What we do **not** invent

- No ARR, revenue, headcount, or labeler counts unless cited.
- No valuation beyond company PR “over $29B” (FACT) and press $14.3B/49% (SECONDARY).
- AWS $1M/mo figures are marketplace **list** dimensions only.

## Teaching arc (INFERENCE from product ladder)

1. **Land:** Self-serve / Rapid labeling + free units → experiment with annotation & data management.  
2. **Expand data:** Managed Data Engine / Generative AI Data Engine + expert workforce for frontier training data.  
3. **Evaluate:** SEAL / Evaluation / red-teaming as trust layer for labs and enterprises.  
4. **Platform:** Scale GenAI Platform — agents over enterprise data with eval flywheel.  
5. **Gov & partners:** Donovan for public sector; CSP marketplaces (AWS); strategic Meta commercial expansion.

## Primary sources

- https://scale.com/
- https://scale.com/pricing
- https://scale.com/genai-platform
- https://scale.com/generative-ai-data-engine
- https://scale.com/donovan
- https://scale.com/blog/scale-ai-announces-next-phase-of-company-evolution
- https://scale.com/blog/leaderboard
- https://api-reference.scale.com/docs/rapid-faq
- https://aws.amazon.com/marketplace/pp/prodview-e7onk7j7ju6cw (Donovan)
- https://aws.amazon.com/marketplace/pp/prodview-3odiiiv22ydjc (GenAI Platform SaaS)