AI data platform for training and validating machine learning models at scale
Scale AI is a data platform that provides training data and model evaluation services for machine learning applications.
AI Panel Score
6 AI reviews
Reviewed
Facts checked against scale.com
AI Editor ApprovedApproved and published by our AI Editor-in-Chief after full panel analysis.Scale AI is a data platform that provides training data and model evaluation services for machine learning applications. The company offers data labeling, annotation, and curation, combining human expertise with automated tools to process text, images, video, and sensor data for model training. Pricing is quote-based, spanning an Enterprise offering and a self-serve Data Engine. Key capabilities include the Data Engine, AI evaluations and benchmarking, government and defense AI applications, and physical AI and robotics data. TopReviewed's six-seat AI review panel scored it 7.6/10, praising Meta's $14.3 billion stake as all but guaranteeing the company's continuity while noting that OpenAI and Google reportedly pulled labeling work after the deal, shifting customer concentration. It best fits enterprises that need labeled training data at frontier-lab quality, with the self-serve Data Engine letting research teams pilot pay-as-you-go before committing to enterprise contracts averaging around $93,000 per year.
Scale AI is a comprehensive data platform designed to accelerate the development and deployment of artificial intelligence applications. The company specializes in providing high-quality training data through a combination of human annotation services and automated data processing tools.
The platform serves enterprises, government agencies, and AI researchers who need labeled datasets for training machine learning models. Scale AI handles various data modalities including computer vision tasks like image and video annotation, natural language processing for text data, and sensor data processing for autonomous systems. Their services cover data collection, labeling, quality assurance, and model evaluation.
Key capabilities include expert-sourced data annotation (Scale says 25% of its contributors hold advanced degrees), automated data processing pipelines, quality control systems, and custom dataset creation. The platform also offers model evaluation services to help organizations assess AI performance and identify areas for improvement. Scale AI operates across multiple industries including autonomous vehicles, robotics, healthcare, energy, and defense.
The company positions itself as an infrastructure provider for AI development, handling the complex and time-intensive process of data preparation so organizations can focus on model development and deployment. Beyond the Data Engine, Scale sells the GenAI Portfolio (SGP, also called the GenAI Platform) for building and running enterprise AI agents and Donovan, an AI-agent product for public-sector missions, and says 90% of the world's leading generative AI model builders are powered by its data.
Identifies enterprise and government AI use cases, builds the system, and owns the deployment outcome.
A research initiative combining frontier research, data, and deployment experience to build AI systems that work in real-world settings.
Runs private leaderboards and expert-level evaluations to benchmark and improve frontier AI model capabilities.
Combines automated tools with human oversight to keep humans involved in AI system decision-making and quality control.
Sources and delivers curated, high-quality training data using contributors with advanced expertise for frontier AI models.
Recruits data contributors with precision, including a significant share holding advanced degrees, to meet frontier AI data quality bars.
Processes diverse data types and use cases across industries such as defense, healthcare, robotics, energy, and logistics to power AI training.
Scale no longer publishes self-serve rates or a free allowance; scale.com/pricing now leads to a demo request. Accounts can still be created at dashboard.scale.com.
Ideal for strategic AI initiatives; custom pricing via demo booking, includes access to Data Engine and GenAI Platform
Ex-Google Cloud COO Francis deSouza now runs a 49%-Meta-owned Scale — operator-led, customer mix just shifted.
“Meta acquired a 49% stake for $14.3 billion in June 2025 and Alexandr Wang departed for their superintelligence lab, with strategy chief Jason Droege running Scale as interim CEO until Francis deSouza took over in August 2026. Revenue was about $870 million in 2024 across the Data Engine, GenAI Platform, and government workloads, but OpenAI and Google reportedly cut labeling ties after the Meta deal.”
Jason Droege took the interim CEO seat in June 2025 after Meta paid $14.3 billion for 49% and Alexandr Wang walked across the street to run their superintelligence lab. Francis deSouza replaced him as permanent CEO in August 2026. Founder-led for nine years, now operator-led. That's not nothing.
Scale generated roughly $870 million in 2024, with the Data Engine pulling frontier labs and the GenAI Platform pulling Fortune 500 deployments. Enterprise contracts average around $93,000 a year, scaling past $400,000 for strategic accounts. Labelbox and Surge AI keep nipping at the labeling layer, but neither has Scale's government and defense surface area.
But the catch is the Meta deal reportedly cost Scale work from OpenAI and Google, who pulled labeling contracts after the announcement. Customer concentration risk just got real. Scope a Data Engine pilot on a research workload through the demo call, watch where enterprise renewals land under the new CEO, then decide.
90% of leading generative AI model builders use Scale data per the vendor — Labelbox and Surge AI trail on government surface area.
The board will ask about Meta entanglement and the reported OpenAI and Google labeling exits in 2025.
Every path, pilots included, starts with a demo call, so time to first labels runs on Scale's sales cadence.
Data Engine and GenAI Platform fit cleanly for enterprises building AI, but Meta alignment narrows neutrality.
Meta's $14.3 billion stake plus government and defense contracts mean Scale exists well past a three-year horizon.
Enterprises who need labeled training data at frontier-lab quality.
Teams who can't tolerate vendor-neutrality concerns post-Meta deal.
Meta's $14.3B for 49% of Scale changes what a 2028 data infrastructure bet actually means.
“Meta took 49% of Scale AI for $14.3 billion in June 2025 and Alexandr Wang left to run Meta's superintelligence team. For a Head of Data Infrastructure picking a labeling substrate through 2028, that ownership structure is now the strategic question, not the feature checklist.”
Meta paid $14.3 billion in June 2025 for 49% of Scale, non-voting, and Alexandr Wang walked over to lead Meta's superintelligence push. Jason Droege ran Scale as interim CEO until Francis deSouza took over in August 2026. For a Head of Data Infrastructure picking a labeling and evaluation substrate through 2028, that ownership chart is the first slide of the diligence pack.
The Data Engine and GenAI Platform are the substrate — contributors sourced with 25% holding advanced degrees, private benchmark leaderboards, and Scale Labs research feeding the loop. There is no public rate card, since every path starts with a demo call; Enterprise averages around $93K per year. Mayo Clinic on the logo wall and a BAE Systems defense partnership say the regulated-domain motion is real.
But Surge AI poached frontier-lab spend after the Meta deal, and Labelbox owns the self-serve lane for ML teams that want tooling without the consulting layer. The catch is whether a 49%-Meta-owned vendor still feels neutral to OpenAI, Anthropic, and Google by 2027.
90% of leading generative model builders use Scale data, though Surge AI and Labelbox now contest the flanks.
Data Engine, RLHF pipelines, and private evaluation leaderboards match how senior data infra leads actually procure.
GenAI Platform plus government compliance work fits enterprise and regulated-domain stacks like Mayo Clinic and BAE Systems.
Meta's 49% non-voting stake creates a multi-year neutrality question with rival labs through 2028.
Scale Labs research plus Data Engine signals frontier-grade craft, not a labeling commodity.
Enterprise AI teams who need labeled data at frontier scale.
ML teams who want self-serve tooling without a consulting layer.
Zero published rates: Scale's pricing link ends at a demo form, so every budget starts on a call.
“Scale publishes no rate card, tier list or free allowance — the pricing link and every Get Started button end at a 1:1 intro-call form. Budget from the reported ~$93K average contract, and price Scale Pro's SLAs into the ask.”
Every route to a Scale price ends at one form. The Pricing link redirects to a demo page. So does every Get Started button. I couldn't find a rate card, tier list or free allowance anywhere. Dashboard signup opens a Labeling account. No unit price quoted.
So budget from outside numbers. The reported average contract is about $93,000 a year, some past $400,000. Three years at $93K = $279K. A planning figure, not a quote.
The docs say what the money buys. Scale Pro: API-initiated labeling, dedicated Engagement Managers, quality guaranteed via SLAs. Enterprise adds the GenAI Platform, deployable in your VPC on AWS, Azure or GCP. Solid scope, but I couldn't find overage, volume breaks or renewal terms to benchmark against the obvious alternative. Francis deSouza, CEO since August 10, 2026, promises "provable outcomes." Put that in the SOW.
Procurement starts with a 1:1 intro call, not a card; Scale Pro includes dedicated Engagement Managers and SLA-backed quality once signed.
Enterprise contracts are custom annual with SLAs; no published termination, renewal, or downgrade terms.
Nothing is priced in public — the pricing link redirects to a demo-booking form, with no tiers, unit rates or free allowance.
90% of leading GenAI model builders use Scale data, with named deployments at Mayo Clinic, Cengage, and BAE Systems.
Avg Enterprise contract reported near $93K/year with some past $400K, but no public model for volume scaling or overage.
Enterprise ML teams who need SLA-backed labeling at production volume.
Teams who need a published unit price before signing.
Scale's task API retries failed callbacks for 24 hours, but project setup leans on an Engagement Manager.
“Scale Pro is built for API-driven labeling pipelines, with callbacks that retry for 24 hours and task creation at 30 requests per second. The catch is the front door: Get Started leads to a demo form, no per-task rate is published, and Enterprise contracts reportedly average $93K a year.”
Callbacks retry up to 20 times over 24 hours when your endpoint misses a 2xx. Small detail. Scale Pro is built for labeling pipelines: tasks go in through the API, results POST back as JSON, and a scale-callback-auth header verifies the sender. Task creation runs at 30 requests per second.
The catch is the front door. Get Started leads to a demo form, the docs indicate Engagement Managers set up custom projects, and I couldn't find a per-task rate anywhere on Scale's site. Enterprise contracts reportedly average $93K a year. That's procurement before the first batch lands.
Once inside, the depth is real. Nucleus handles curation and annotation review, lidar tasks get their own API reference, and Scale says 25% of contributors hold advanced degrees. Docs read like the platform team writes them: retry rules, a limits table, a compatibility policy. But Chrome is the only fully supported browser.
Once a project exists, tasks flow through the API with callbacks, but setup leans on Engagement Managers and no page states what a task costs.
Callback retry rules, a technical-limits table and an API compatibility policy read like the platform team owns the docs, not marketing.
Reported ~$93K/year contracts, no published per-task rate, and Chrome as the only fully supported browser add up across a working week.
RLHF, sensor data, model evals, and private benchmark leaderboards stack into genuinely deep advanced surface for ML teams.
Data Engine plus GenAI Platform cover the MLE pipeline end-to-end, though enterprise gating breaks the casual upload-and-label loop.
ML engineers who feed labeling tasks from their own pipelines.
Solo developers who need a per-task price before starting.
Scale AI still opens labeling accounts, but every Get Started button points you to an intro call.
“Scale Pro and Nucleus are serious daily tools, with API-started labeling, SLAs and annotation review. Getting in is the awkward part, since the pricing page now redirects to a demo form and I couldn't find a free allowance.”
"Get Started" on Scale's homepage really means "book a 1:1 intro call." Every one lands on the same demo form, and scale.com/pricing now redirects there too. The signup at dashboard.scale.com still opens labeling and Nucleus accounts, but none of the Get Started buttons lead there. No free allowance I could find, and no per-unit rate. Enterprise reportedly averages around $93K a year.
Past that door, the tooling is serious. Scale Pro starts labeling from the API, handles 3D and sensor-fusion data, and backs quality with SLAs. Nucleus is where month three gets interesting: curate the ugly slices, review annotations, measure the model against them.
The catch is the front door. Dedicated Engagement Managers are great if you want a human on the hook, but a small team that wants labels tonight has to hunt for a signup no Get Started button points to. Powerful pipes. Unmarked door.
The public site is tidy, but "Get Started" really means "book a call," and the dashboard's day-to-day feel isn't visible from outside.
Data Engine, Nucleus, Scale Pro and the GenAI Platform each get their own docs section, so month-three discovery is real but signposted work.
Data labeling is a desk job; mobile parity is not the meaningful axis for this category.
A signup exists at dashboard.scale.com, but every Get Started button routes to an intro-call form and I couldn't find a free allowance.
Scale Pro docs guarantee label quality via SLAs, and Meta, Mayo Clinic, BP and CDAO on the homepage signal a pipeline built for serious load.
ML teams who want a dedicated Engagement Manager running their labeling projects.
Small teams who want published per-unit prices before they label anything.
Meta took 49% for $14.3B in June 2025 — and the customers Scale relied on started leaving.
“Meta paid $14.3B for 49% of Scale in June 2025, Alexandr Wang left to run their superintelligence lab, and Google plus OpenAI started pulling work. Then 200 layoffs in July — 14% of staff, official line: GenAI capacity ramped up 'too quickly'.”
The whole bet just changed. Meta took 49% for $14.3B in June 2025, Alexandr Wang walked to run their superintelligence lab, and Google, OpenAI, and Microsoft started pulling work. Customer concentration risk firing in real time.
What's left: the Data Engine repackaged for enterprise and government — Mayo Clinic, BAE Systems, the CDAO. Interim CEO Jason Droege ran Uber Eats, and Francis deSouza took the permanent seat in August 2026. Decent operators, but two CEO changes in fourteen months. But labor-arbitrage labeling has Surge AI, Labelbox, and Snorkel swarming the same shops.
Then 200 layoffs in July 2025, roughly 14% of staff. Official line: GenAI capacity was ramped up 'too quickly'. The catch — Scale must rebuild a base its primary investor now competes with. Enterprise deals average $93K. Pricing is fine. The thesis isn't.
Surge AI, Labelbox, and Snorkel cover the same labeling shops, and Meta's 49% actively repels frontier-lab buyers Scale once owned.
Labeled data and evaluation outputs are portable artifacts customers own — migration off is mostly a contract decision.
Survives via enterprise and government pivot with Meta balance-sheet backing, but 200 July 2025 layoffs and customer flight muddy a 3-year bet.
Landing copy ("Reliable AI for the world's most important decisions") is less hype-laden than peers, though the 90%-of-frontier-builders claim now reads stale post-Meta.
Pattern matches once-hot infra plays whose customers churned after strategic-investor deals — early signal, not yet conclusion.
Enterprise and government teams who need vetted training data at SLA.
AI labs who compete with Meta on frontier models.
Common questions answered by our AI research team
Scale AI processes text, images, video, and sensor data for machine learning training, combining human expertise with automated tools to prepare this data for AI model development.
Yes. Beyond data labeling, Scale AI offers the Scale GenAI Portfolio (SGP, also called the Scale GenAI Platform) for building and running enterprise AI agents, and Scale Donovan for deploying AI agents in mission-critical public-sector workflows, along with its core Scale data engine product.
Scale AI serves a wide range of industries including life science, medicine, energy, infrastructure, defense, healthcare, logistics, robotics, autonomy, real estate, and government/public sector operations.
Yes. Scale AI powers 90% of the world's leading generative AI model builders, including a partnership with Meta to accelerate its LLM and generative AI development.
You can get started with Scale AI by booking a demo directly through the website to see how Scale can build reliable AI for your organization.
Who Scale AI competes with, and how each one differs
Runs a comparable RLHF/data-annotation workforce (~1M annotators) but has stayed bootstrapped with no outside investors, reporting $1.2B in 2024 revenue and picking up work from Scale's own customers (OpenAI, Google, xAI) after Meta took its stake in Scale.
Pairs a self-serve annotation platform with the Alignerr expert network (2.6M+ contributors across 200+ domains), targeting frontier labs with domain-specialist training signals rather than Scale's larger generalist annotator workforce.
Runs an expert-labor marketplace paying an average $113/hour across 300+ professional fields (law, medicine, engineering) plus its own APEX model benchmarks, competing for frontier-lab budgets with a smaller, higher-skill labor pool than Scale's broader crowd.
A publicly traded (ASX: APX) crowdsourced annotation vendor with roughly 1M workers across 130+ countries and $273M in 2023 revenue, competing on price and language coverage rather than Scale's frontier-AI and defense focus.
Takes the opposite approach to Scale's human-in-the-loop model: Snorkel's programmatic labeling generates training data from code-defined rules rather than annotator workforces, competing for the same enterprise data-preparation budgets.
Scale delivers proven data, evaluations, and outcomes to AI labs, governments, and the Fortune 500.
