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

At a glance#

What each company does#

Scale AI#

Scale AI was founded in 2016.¹¹ It started as a data labeling provider for machine learning teams and, according to TechCrunch in 2024, offered human-in-the-loop labeling and industry datasets such as camera and lidar data for self-driving cars.¹¹ Alexandr Wang, its founder, is now chairman of the board.²

Today Scale describes three areas on its website: a data engine for model builders, a GenAI portfolio for enterprises, and a product called Donovan.¹ It says 90% of leading generative AI model builders are "powered by" Scale.¹ The data engine page describes post-training work such as prompt-response generation, RLHF (reinforcement learning from human feedback), red teaming and evaluation, plus annotation of text, image, video and 3D sensor fusion data.⁶

The enterprise side is a separate pitch. Scale says it turns models into reliable systems by embedding experts and deploying its Generative AI Platform in a client's environment.¹⁵ In its July 2026 CEO announcement, Scale named BP and Mayo Clinic as recent enterprise wins and said it expanded work with U.S. and international governments, including the U.S. Department of War.²

Surge AI#

Surge AI was incorporated in 2020 by Edwin Chen, a former Twitter, Google and Facebook engineer, according to Forbes.⁸ Forbes reported that Chen started the company with "a couple million" in personal savings and avoided venture capital, and that, by Chen's account, the company has been profitable from close to its first day.⁸

Surge's website describes its offering as new frontier data and RL environments "off the shelf," post-training runs, an expert workforce, enterprise solutions, and a set of public benchmarks.³ Its enterprise page frames the work as three tracks: evaluate, improve and train, and says pilot results can come within a week.⁴ Its blog lists benchmark releases such as GDP.pdf, the Tuesday Work Index, DAYJOB and HANDBOOK.md, and posts about work with Anthropic and Microsoft.¹⁰

History and leadership#

Scale AI: from labeling startup to Meta-backed company#

Scale raised a $1 billion Series F in May 2024 at a $13.8 billion valuation, led by Accel and including Amazon and Meta among new investors, roughly double its 2021 valuation of about $7 billion.¹¹

In June 2025, Meta agreed to invest $14.3 billion for a 49% stake without voting power, and Wang left to join Meta while staying on Scale's board as a director.¹²¹⁶ TechCrunch reported that the investment values Scale at $29 billion, and described the $14.3 billion figure as reported rather than confirmed by the company.¹² Chief strategy officer Jason Droege became interim CEO.¹²

Scale said the deal would not change how it treats customers. CNBC, via NBC Chicago, reported a Scale spokesperson's statement that Meta would not be privy to its business information or data.¹⁶ TechCrunch reported that a Scale blog post sought to dispel the idea that Meta would receive preferential treatment.¹⁷

On July 30, 2026, Scale announced that Francis deSouza, formerly Google Cloud's chief operating officer and earlier CEO of Illumina, would become CEO effective August 10, 2026, with Droege assisting on the transition.² The announcement describes Wang as "Founder and Chairman of the Board."²

Surge AI: founder-controlled and quiet#

Surge discloses little. Forbes reported that Surge operates under nondisclosure agreements, does not explain how it matches workers to projects, and gives customers a link to a finished dataset.⁸ The same profile said Chen has said he is not interested in selling the company and has no IPO plans, and that he owns about 75%.⁸

This page uses Forbes' 2020 incorporation date.⁸

Funding and valuation#

This is the area where public information is thinnest and most conflicting.

A few cautions apply.

• Scale's 2026 revenue figure is an expectation reported by Forbes México, not audited revenue.
• Surge's raise has been described as "in talks" since July 2025. No closed round appears in the sources reviewed here, though one could have occurred since.
• Third-party data sites list different Surge valuations. This page relies on the dated press reports above rather than those aggregators.

Forbes said Scale has roughly four times as many staff as Surge and less revenue, which is one reason the two firms are often compared.⁸ Forbes reported Surge has about 250 employees including part-time staff and consultants.⁸

Business model#

Scale AI has two broad lines. The first is data and evaluation work for AI labs, which was the core business. The second is an applications business that builds custom AI applications for governments and enterprises, which interim CEO Droege said in June 2025 Scale would "double down" on.¹⁷

Customer concentration was a real issue. Reuters reported in June 2025, as relayed by TechCrunch, that Google planned to cut ties with Scale after the Meta deal; Google had planned to pay Scale about $200 million that year, and Scale said its business remained strong.¹⁹ OpenAI separately said it was phasing out work with Scale and that the wind-down began before the Meta announcement.¹⁷

In July 2025 Scale cut about 14% of staff, around 200 people, and ended relationships with about 500 contractors. TechCrunch reported that Droege's memo said the company had scaled its core data-labeling business too quickly.²⁰ The government line has grown: Scale said on May 6, 2026 that the Pentagon's Chief Digital and Artificial Intelligence Office raised the total potential value of its enterprise agreement from $100 million to $500 million.²¹

Surge AI sells expert-labeled data, evaluations and RL environments, mainly to AI labs. Forbes reported that Surge charges 50% to ten times more than competitors, according to two researchers it cited. Forbes also said Surge's customers are concentrated and some have moved work in-house.⁸ Forbes reported that Cohere moved most annotation in-house and that OpenAI confirmed it no longer works with Surge, while Meta still uses Surge.⁸

Forbes said Google's contract with Surge grew to more than $100 million per year.⁸ Surge's own website lists no customers.³ Customer lists for Surge therefore rest on press reports.

How they source data and people#

This is the heart of the contrast, and it matters for anyone thinking about what these companies actually buy.

Scale AI: Remotasks and Outlier#

Scale operates the Remotasks and Outlier contributor platforms.²²⁷ Outlier's website says it is operated by Scale and recruits people with domain knowledge, preferring master's and PhD holders. It lists tasks such as writing hard prompts with correct answers, building grading rubrics and ranking answers.⁷ The page claims 900,000 or more experts and more than $500 million paid to them; these are company figures.⁷ It describes onboarding of 30 to 90 minutes, weekly pay, and no minimum hours, and it does not describe contractor classification.⁷

Remotasks has a record of criticism. A 2023 Washington Post investigation, republished by Stars and Stripes, found that of 36 current and former freelancers interviewed, all but two said payments had been delayed, reduced or canceled, and that Philippine workers often earned below the local minimum wage. Scale said its pay system was improving and that payment delays were rare.²³ Rest of World reported in March 2024 that Remotasks had shut down in Kenya, Nigeria and Pakistan and stopped new sign-ups in Thailand, Vietnam and Poland, which Scale attributed to enhanced security protocols.²²

Surge AI: DataAnnotation and a consultant network#

Surge hires through DataAnnotation Tech, which Forbes called a wholly owned affiliate and noted that job listings do not name Surge.⁸ The DataAnnotation site describes remote, task-based contractor work with no long-term commitment and advertised maximum rates by role.⁹ Forbes reported more than one million gig workers in over 50 countries.⁸ Surge also recruits professionals such as doctors, lawyers and finance experts through its "Expert Workforce" and "Consultant Network."³²⁴

On pay, Forbes reported at least $20 per hour for general work and more than $40 per hour for specialized tasks.⁸ The DataAnnotation page reviewed for this article lists upper bounds: up to $40 per hour for language specialists, up to $50 for generalists, up to $125 or more for many expert roles, and up to $150 or more for software and data science roles. It says rates are the maximum and pay depends on output quality.⁹ These are not average earnings.

Forbes also reported that Surge monitors annotators with hidden quizzes, manual reviews and machine learning systems.⁸

Side by side#

Legal issues and controversies#

The allegations below are unproven. Each company disputes them.

Scale AI#

• Misclassification suits. TechCrunch reported in March 2025 that two former workers had sued, in December 2024 and January 2025, alleging underpayment and misclassification as contractors. Scale said it complies with the law and that its pay meets or exceeds local living wage standards.²⁵
• Department of Labor probe. TechCrunch reported the Department of Labor was investigating Scale's compliance with the Fair Labor Standards Act, active since at least August 2024.²⁵ On May 9, 2025, TechCrunch reported that the DOL had dropped the investigation, according to a source, with no clear reason given. Scale said it was pleased with the update.²⁷
• Psychological harm suit. A federal class action filed January 17, 2025 by six Outlier workers in the Northern District of California alleged inadequate protection from disturbing content. Scale said it has safeguards, including opt-out and advance notice, and would defend itself.²⁸
• Exposed documents. Business Insider reported in June 2025 that Scale used public Google Docs for client projects, some accessible to anyone with a link and some marked confidential. The New York Post's account of that reporting said Business Insider viewed 85 documents, that there was no indication the files had led to a breach, and that Scale said it was investigating and had blocked public sharing from its managed systems.²⁹
• Customer departures. The OpenAI and Google reports above followed Meta's investment.¹⁷¹⁹

Surge AI#

• Misclassification suit. A proposed class action was filed on May 20, 2025 in San Francisco Superior Court by a Surge worker represented by Clarkson Law Firm. It alleges Surge deliberately misclassified annotators and required off-the-clock training.²⁶³⁰ Forbes reported that Chen said the suit is without merit. The sources reviewed show no later ruling or settlement.⁸
• Leaked documents. Business Insider obtained a Surge spreadsheet listing websites workers could use or avoid while training Anthropic's models. Tom's Guide reported that Anthropic said it had no knowledge of the list, which was reportedly created independently by its vendor, and that Surge took the document offline after the leak was reported.³¹
• China-related reporting. In an August 2026 Forbes investigation (read here in its Spanish-language edition, Forbes México), a message seen by Forbes quoted a Tencent data-procurement executive saying the company relies on Surge, and the story said Surge's CEO traveled to China to meet lab executives. Surge said it does not disclose client information, Tencent and other Chinese labs did not respond to requests for comment, and the article did not allege that selling training data to Chinese labs is illegal.⁵ Treat this as a press report, not an established fact.
• Pricing and transparency. Forbes quoted Chen calling Scale "dead in the water."⁸

Strengths and weaknesses#

Scale AI

• Strengths: the longest track record, a government and defense business, an enterprise deployment business, and a published program that pays companies for de-identified operational data (see below).^1152132
• Weaknesses: reliance on a few large customers, which wavered after the Meta deal; a layoff in 2025; labor and security controversies; and a leadership change in 2026.^1719202

Surge AI

• Strengths: founder-stated profitability, a large reported revenue figure for its staff size, public benchmarks cited by labs, and founder control.⁸¹⁰
• Weaknesses: limited disclosure, a concentrated customer base, some customers moving work in-house, a pending misclassification suit, and China-related scrutiny.⁸²⁶⁵

Who each suits#

• AI labs that want breadth and scale across data, evaluation and applications may prefer Scale, which sells all three. Labs wary of Scale's Meta ties may weigh that differently; Scale has said Meta gets no preferential treatment.¹⁷
• Labs that want expert-written data, RL environments or benchmarks from a smaller specialist may prefer Surge, whose public output centers on benchmarks and environments.¹⁰
• Governments are an explicit Scale segment.¹² Surge's U.S. Army and Air Force customers appear in press reports, not on Surge's own site.⁵
• Enterprises can use either: Scale through its deployed-applications model, Surge through its evaluate, improve and train programs.¹⁵⁴
• Contributors should read the contractor terms, and know that both firms are defendants in misclassification suits.

What this means for companies that want to sell or license data#

Three practical points follow from the sources.

1. Scale's program targets operational data. The data types it lists include CRM, chat, email, documents and project tools (see CRM history and communications data).³²
2. De-identification is part of the pitch. Scale says it removes personal information inside the client's environment before extraction and uses per-package approval.³² A seller should still confirm what that means contractually.
3. Expert services are a different path. Surge and Outlier recruit individuals. A business with domain experts is not the same as a business with data to sell, and the pay figures above are for contractor labor.

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.

Summary#

Scale AI and Surge AI both supply the human-made data that AI labs use to train and test models, but they are shaped differently. Scale is older, has more staff and is more diversified, and Meta holds 49% of it without voting power; it faced customer pullbacks and layoffs in 2025, has expanded into government and enterprise work, and announced a new CEO in July 2026, effective August 10. Surge is founder-owned, says it has been profitable, and focused on expert data, benchmarks and RL environments, but discloses little. Both face misclassification lawsuits that they dispute. For data sellers, Scale is the only one of the two with a published program to pay companies for de-identified business data.

Key takeaways#

• Scale is broader (data, evaluation, applications); Surge is narrower (expert data, benchmarks, RL environments).
• Ownership and funding differ sharply: Meta's 49% non-voting stake in Scale versus a founder-held Surge with reported but unconfirmed fundraising talks.
• Reported figures conflict (Surge valuation talks, revenue comparisons), so check dates and sources.
• Both depend on contractors and face California misclassification suits; Scale also had a DOL probe that ended in May 2025.
• Data sellers should look to Scale's published data partnership terms, not to Surge, and should verify terms directly with any buyer.

Verdict#

There is no single winner, because the two companies solve different problems. Scale AI fits organizations that want a large vendor spanning data, evaluation and deployed applications, including governments and big enterprises, and operating companies whose workflow records match its data partnership criteria. Surge AI fits frontier labs and enterprises that want expert-written data, custom benchmarks and RL environments from a founder-controlled specialist, with the caveat that much of what is known about it comes from press reports. As of October 2026, buyers and sellers should recheck funding, leadership and litigation before deciding, since each has changed within the past year.