Scale AI vs Labelbox: How They Compare for AI Training Data (2026)

Summary#

Scale AI and Labelbox sit in the same part of the AI supply chain. Both help AI developers turn human judgment into training and evaluation data. They differ in history, ownership, size and emphasis. Scale AI, founded in 2016, has long sold managed data services and, as of October 2026, describes a push into enterprise and government applications.¹²³ Labelbox, founded in 2018, started as a software platform for labeling and has added managed labeling, reinforcement learning environments and an expert network called Alignerr.⁴⁵⁶

Two events shape the comparison. In June 2025, Meta invested $14.3 billion for a 49% stake in Scale with no voting power, valuing Scale at about $29 billion; Alexandr Wang left for Meta and Jason Droege became interim CEO.⁷⁸⁹ On July 30, 2026, Scale announced that Francis deSouza would become CEO effective August 10.¹⁰ On October 14, 2026, Labelbox announced more than $100 million in Series E financing, including equity and a credit facility.¹¹ This page compares what each company does, how each sources people and data, what is publicly known about money and legal matters, and what it all means for a business that is thinking about selling or licensing its own data.

At a glance#

Customer lists on this table come from each company's own web pages and press, not from independent confirmation. Funding figures are cited in the sections below, and where sources conflict the conflict is noted.

What each company does#

Scale AI#

Scale describes itself as a builder of AI systems for enterprises and governments, covering the stack from training data to the systems that deploy models.² Its Scale Data Engine page defines the data engine as improving machine learning models with high-quality, diverse and large datasets "powered by experts," and lists work such as prompt-response pairs, human preference data for reinforcement learning from human feedback (RLHF), red teaming and evaluation.¹²

The company's other named products as of October 2026 are the Scale GenAI Portfolio and Scale Donovan.² Scale's documentation describes its GenAI Platform as a way for enterprises to develop, test and deploy generative AI applications with their own data, including fine-tuning, retrieval-augmented generation pipelines and evaluation tools, with applications launched securely in the customer's VPC on AWS or Azure.¹⁹ Donovan is presented as a platform for building, testing and deploying specialized AI agents, aimed mainly at defense and government users, and the page says it is available on classified and air-gapped networks and lists DISA IL4 and FedRAMP High authorization for its unclassified GovCloud environment.²⁰ These are vendor claims; the page does not offer independent verification.

Scale also runs research and evaluation work. Its January 2026 strategy post says the SEAL lab introduced 15 new benchmarks and published more than 450 evaluations across more than 50 models in 2025.³

Labelbox#

Labelbox began as a software platform. Its original suite, as described in its documentation and a Google Cloud architecture guide, includes Catalog (search and organization of unstructured data), Annotate (labeling with review and workforce management, usable with an internal team, another vendor or Labelbox's own services), Model (model diagnostics) and Boost (managed access to specialized labeling services).²¹²² The current Labelbox homepage emphasizes newer products: Horizon (reinforcement learning environments and evaluations for post-training), Recursion (agents that run in the background on triggers), Terra (robotics data) and Alignerr (an expert network).⁵ The homepage as reviewed did not mention Catalog, Annotate, Model or Boost, so it is unclear whether the company still presents them as separate lines.

Labelbox says hundreds of AI teams build with it and that over 90% of leading US AI labs post-train and evaluate on it.⁵ Scale makes a comparable claim about frontier model builders: "90% of the world's leading generative AI model builders" are powered by Scale.² Both claims are self-reported, and the two companies measure different things, so they should not be compared directly.

History and founding#

Scale was founded in 2016. TechCrunch's 2024 funding report calls it a data-labeling startup and names Alexandr Wang as CEO and co-founder.¹ Scale has since added RLHF, LLM evaluation and enterprise software, and in 2026 describes itself as a builder of AI systems for enterprises and governments.²¹²

Labelbox was founded in 2018. Labelbox's own About page gives 2018 as the founding year and lists Manu Sharma as co-founder and CEO.⁴ A Cerebral Valley profile says Sharma and two co-founders started it after seeing at Planet Labs that deep learning models needed high-quality human-labeled data, built the product on nights and weekends, and launched it on Reddit.⁶ In its January 2022 Series D post, Labelbox said enterprises produced 80% of its 2021 business and named customers such as Genentech, Warner Bros, Black & Decker and Stryker.¹³

Funding and valuation#

Scale AI#

In May 2024, Scale raised a $1 billion Series F, a mix of primary and secondary funding, led by Accel at a $13.8 billion valuation, roughly double its 2021 valuation of about $7 billion. New investors included Amazon, Meta and the venture arms of Cisco, Intel, AMD and ServiceNow.¹

In June 2025, Meta invested $14.3 billion for a 49% stake with no voting power, valuing Scale at about $29 billion. TechCrunch reported the $14.3 billion, the 49% stake and the $29 billion valuation, and CNBC reported that a Scale spokesperson said Meta would have no voting power.⁷⁸ Scale's own announcement did not state the investment amount; it said Meta would hold a minority of Scale's outstanding equity, that the investment values Scale at over $29 billion, that Alexandr Wang would remain a director, and that Chief Strategy Officer Jason Droege would be interim CEO.⁹ Scale also said it would distribute proceeds to shareholders and vested equity holders, giving them substantial liquidity.²³

Scale's revenue figures are not audited or filed publicly. Its January 2026 post says 2025 was its biggest revenue year and that it won well over $1 billion in new business, but it gives no revenue total.³

Labelbox#

Labelbox announced a $110 million Series D on January 6, 2022, led by SoftBank's Vision Fund II, with Databricks Ventures, Snowpoint Ventures and previous investors B Capital, Andreessen Horowitz and Catherine Wood of ARK Invest.¹³

On October 14, 2026, Labelbox announced more than $100 million in Series E financing, including equity and a credit facility, at a $1.25 billion pre-money valuation. Infinitum Partners led, with Wipro Ventures, a16z, B Capital, Tau Ventures, Snowpoint Ventures and others participating. The company said the money would go to research and engineering for Horizon and Recursion, hiring, and securing compute access.¹¹ Because the round includes a credit facility, the full amount should not be read as new equity.

The same post carries company claims: 400+ customers, profitability reached this year, and growth of more than 5x year over year. These figures are unaudited. The post also says Recursion Managed Agents opened to every team the week before.¹¹

Business model#

The core difference is the unit of sale.

Scale has historically sold managed data services: a customer describes what it needs, and Scale supplies the people, process and quality control. Its January 2026 strategy post says the data business "grew strongly," performed best in the second half of 2025, and is now profitable, and that Scale plans to move technical capabilities from the data business into applications faster.³ It also says applications revenue more than doubled in the second half of 2025 and is expected to roughly double again in 2026.³ On July 30, 2026, Scale announced that its board appointed Francis deSouza (formerly Google Cloud's chief operating officer and Illumina's CEO) as CEO effective August 10, 2026, succeeding interim CEO Jason Droege. The announcement describes Alexandr Wang as founder and chairman of the board and names BP and Mayo Clinic as recent enterprise wins.¹⁰

Labelbox sells software first. Customers can use the platform with their own labeling team, another vendor, or Labelbox's services.²¹ A Cerebral Valley profile says Labelbox also offers managed labeling and an arrangement where customers pay only when output meets set quality thresholds.⁶ In November 2024, it launched Alignerr Connect, which lets companies hire vetted experts directly and run them in their own tools, and described it as a complement to the platform and managed services.²⁴ The company claims to be the only vendor offering platform, managed services and direct hiring in one suite.²⁴ That is a marketing claim.

In practice the lines have blurred. Scale sells software and applications as well as services, and Labelbox sells services and a workforce as well as software. The gap now is mainly emphasis: Scale is publicly moving toward deployed applications for large buyers, and Labelbox is leaning into reinforcement learning environments and expert-generated evaluation data for labs.³¹¹

Who they serve#

Scale lists enterprise, insurance, healthcare, national security and public sector segments on its site, and names customers including NVIDIA, Mayo Clinic, BP, Meta and TIME.² Its January 2026 post says the US government business had its best quarter in Q3 2025, that the Department of War awarded two contracts totaling nearly $200 million in 2025, and that Scale leads Project Thunderforge with Anduril and Microsoft.³ In April 2026, Scale announced it had acquired ICG Solutions, a defense technology company focused on real-time streaming data analytics, with terms not disclosed.²⁵

Labelbox's homepage shows case studies with Meta, Google Cloud, Intuitive Surgical and an unnamed retailer, and logos for companies such as Ideogram, Shutterstock, Pinterest, Walmart and Etsy.⁵ A Cerebral Valley profile says Google Cloud offers a human LLM evaluation service powered by Labelbox, and John Deere uses a Labelbox-powered data engine.⁶ Logos on a vendor website show a relationship but not its size or recency.

Customer concentration after the Meta deal#

Customer reaction to Meta's stake is the most consequential public event for Scale. TechCrunch, citing Reuters, reported in June 2025 that Google, which had reportedly planned to pay Scale about $200 million that year, was planning to cut ties and talk to competitors, and that Microsoft was also reportedly looking to pull back. A Scale spokesperson said the business remained strong and that Scale would keep operating independently and safeguarding customer data. The same report said OpenAI had reportedly made a similar decision months earlier, although OpenAI's CFO said it would keep working with Scale as one of many vendors.²⁶

On July 16, 2025, TechCrunch reported, citing Bloomberg, that Scale was cutting about 14% of its staff (roughly 200 employees) and ending work with 500 global contractors. According to the report, interim CEO Droege's memo said Scale had scaled its core data-labeling business too quickly, and the company planned to hire in enterprise and government sales.²⁷ Scale's January 2026 post nonetheless describes 2025 as its biggest revenue year and says it won well over $1 billion in new business.³ The sources do not fully reconcile these accounts.

How they source data and people#

Scale AI#

Scale runs separate contributor platforms. Its own overview describes Remotasks as a global platform for flexible, gig-based data annotation, kept separate from the main platform to protect customer confidentiality.¹⁴ Outlier describes itself as a platform operated by Scale AI that connects experts with leading AI companies.¹⁶ Scale's site says about a quarter of its contributors hold advanced degrees.²

For higher-skill work, Scale's affiliate hires directly. A posting on Y Combinator's job board for a PhD data partner role lists $50 per hour, a minimum of three weeks and a requirement to hold a completed PhD; the contracting party is Smart Ecosystem, Inc., which the posting calls a Scale AI affiliate.¹⁵

Labelbox#

Labelbox sources experts through Alignerr, which says it is "powered by Labelbox" and hires "PhDs, researchers, subject matter experts, and experienced professionals across disciplines."¹⁸ Alignerr's headline figures conflict. A February 2026 Labelbox post described a network of "over 1 million domain experts," while the Alignerr homepage reviewed in October 2026 says "Join 100,000+ Experts Earning by Training AI."¹⁷¹⁸ The two may define "experts" differently; neither figure is independently verified.

In February 2026, Labelbox announced it had acquired Upcraft, whose team it said would help automate expert recruitment and engagement for Alignerr. Terms were not disclosed.¹⁷ Alignerr Connect says candidates are screened using interview transcripts, assessment scores and past project metrics, and claims access to the "top 3%" of data talent.²⁴ The "top 3%" figure is unverified marketing.

Contractor pay and terms#

Pay is only partly public, and what is public is mostly advertised rates.

• Scale (Remotasks): Scale's overview says it works with the Global Living Wage Coalition, runs quarterly pay analyses and cites a survey of more than 5,000 workers in which nearly 75% gave a satisfaction score of 4 or 5 out of 5. The page gives no pay rates and no survey methodology.¹⁴
• Scale (affiliate contract role): $50 per hour for a PhD data partner on a minimum three-week contract.¹⁵
• Scale (court allegations): A class action complaint in San Francisco Superior Court alleges Outlier workers were recruited with promises of $25 to $40 per hour and that pay was reduced or denied through algorithmic rules. These are allegations, not findings.²⁸
• Alignerr: Its site lists $40 to $120 per hour for most remote roles, $25 to $30 for Italian language expert roles and $20 to $40 for Japanese localization roles, and claims an $80 per hour average. The page does not explain how that average is calculated.¹⁸

Neither company's public pages reviewed here spell out full contract terms, such as minimum hours or termination. Alignerr's help center only says that labelers use Labelbox with issued credentials and that some annotations are peer reviewed.²⁹

Controversies and legal matters#

These items are reported neutrally. Allegations are not proven findings.

• Worker classification suits against Scale. A class action filed by former worker Steve McKinney alleges Scale misclassified Outlier workers under California's ABC test and seeks to represent an estimated 10,000 to 20,000 California workers.²⁸ TechCrunch reports that ex-workers have sued Scale alleging they were underpaid and misclassified as contractors, and that a Scale spokesperson said the company is proud of the "flexible earning opportunities" its platforms offer.³⁰
• Department of Labor. TechCrunch reported in March 2025 that the Department of Labor was investigating Scale's Fair Labor Standards Act compliance, and in May 2025 that the probe had been dropped, with the reason unclear (citing a source familiar with the matter). Scale said it was "pleased with this update."³⁰
• Labelbox. The sources reviewed for this page did not surface lawsuits or regulatory actions against Labelbox. That is a limited search result, not a clean bill of health.

Recent news through October 2026#

• February 10, 2026: Labelbox announced the Upcraft acquisition.¹⁷
• April 20, 2026: Scale announced the ICG Solutions acquisition.²⁵
• July 30, 2026: Scale announced Francis deSouza as CEO, effective August 10.¹⁰
• October 14, 2026: Labelbox announced its Series E and said Recursion Managed Agents had opened to every team the week before.¹¹

Strengths and weaknesses#

Scale AI#

Strengths: longer track record, broad service range from data to deployed applications, large government and enterprise relationships, published research and benchmarks, and Meta's capital.³⁷ Weaknesses: public reporting of customer pullback after the Meta deal, the July 2025 layoffs, contractor litigation, leadership turnover (Wang to Meta, an interim CEO, then a new CEO in 2026), and a stated pivot that may stretch focus.²⁶²⁷¹⁰

Labelbox#

Strengths: software-first approach that lets customers keep control of their own process, a spread of service options, a large claimed expert network, fresh capital and company-reported profitability.¹¹²¹²⁴ Weaknesses: smaller scale and less public financial detail, company-reported figures (400+ customers, profitability) that are unaudited, expert-network figures that are not independently verified, and a product line that has changed enough that older descriptions of its tools may be dated.¹¹⁵

Who each suits#

Scale AI fits organizations that want a single provider to run data work and also help build deployed AI systems, especially in defense, government and large enterprise settings. Buyers should weigh Meta's ownership stake when the data in question is sensitive, noting that a Scale spokesperson said Meta would not have access to Scale's business information or data.⁸

Labelbox fits teams that want to run their own labeling or evaluation process on a platform, with the option to add managed services or direct-hire experts. AI labs focused on reinforcement learning environments and expert-written evaluations are its stated focus.¹¹⁵

The data-seller angle#

A few practical points follow from the comparison:

1. Role matters. An AI lab that wants training data may hire Scale or Labelbox to create it. That is a different relationship from a lab buying a finished dataset.
2. Ownership and access. Scale told reporters that Meta would not have access to its business information or data and that it is committed to safeguarding customer data.⁸²⁶ Sellers who care about which labs can see their material should ask any intermediary for written commitments.
3. Expert labor is a cost center for these firms. Both companies recruit contractors and advertise hourly rates, which is a model built on paying people, not licensing assets.¹⁵¹⁸

Do not let one buyer set your price. Any single buyer's published range, program page or first offer is that buyer's number, not the market's. The sensible move is to get competing offers on the same data, compare scope, exclusivity, indemnity and payment timing side by side, and have someone independent look at the terms. Our guide to selling direct vs using a broker explains how, and if a buyer is already talking to you, we review offers free.

Verdict#

Neither company is simply "better." Scale AI is the larger, more established firm and is repositioning toward enterprise and government applications, while carrying the effects of the Meta transaction, litigation over contractor classification, and a leadership transition.³¹⁰²⁸ Labelbox is the smaller platform-and-network company, with a software-first model and a fresh Series E, but with less independently verified detail about its scale.¹¹⁵ The right choice depends on whether a buyer wants a full-service provider or a platform with optional services.

Sources#