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

At a glance#

Figures are as of October 2026. Revenue, employee counts and customer claims are company statements unless a news outlet is named.

Why this comparison matters#

Scale AI defined the commercial market for human data labeling over the last decade. Invisible Technologies is a smaller, quieter company that supplied human feedback data to OpenAI and others while also running an enterprise business.³⁸ In 2025 the two were pulled into the same story: after Meta's investment in Scale, OpenAI and Google DeepMind reportedly cut ties with Scale, and vendors including Mercor grew into the gap.¹⁷

For a business owner, the comparison is useful in three ways. It shows how much of the market sits with a few large vendors. It shows that the model has two shapes: a very large data and evaluation business that is moving into applications, and a mid-sized company that blends training data with enterprise software. And it shows how contested contractor treatment and customer neutrality have become. This page summarizes what each company says about itself and what independent outlets report, and flags where the record is thin.

Scale AI#

History and funding#

Scale says its work dates to 2016 and describes itself as a provider of reliable AI systems for high-stakes decisions.³ In May 2024 it raised a $1 billion Series F at a $13.8 billion valuation, led by Accel, a round TechCrunch described as a mix of primary and secondary funding that roughly doubled the valuation from its 2021 round of about $7 billion. Amazon and Meta were among the investors. At that point TechCrunch listed Microsoft, Toyota, General Motors, Meta, the US Department of Defense and OpenAI among its customers.¹²

Meta's 2025 investment#

In June 2025 Meta invested $14.3 billion for a 49% stake in Scale with no voting power, according to a Scale spokesperson quoted by CNBC.¹⁸ Scale said the deal valued it at about $29 billion, and Alexandr Wang left as CEO to join Meta while staying on Scale's board as a director. Chief strategy officer Jason Droege became interim CEO.¹³ The $14.3 billion figure was first reported by Reuters from two sources; Scale itself did not state an amount.¹³

In a June 18, 2025 memo, Droege told employees that Scale remains an independent company. In the same CNBC report, OpenAI said it had been winding down work with Scale over the previous six to 12 months and that the Meta deal did not influence the decision, and Google was reported to be planning to cut ties.¹⁹ TechCrunch, citing Reuters, reported that Google had planned to pay Scale about $200 million that year and was talking to competitors, while OpenAI's CFO said it would keep working with Scale as one of many vendors.²⁰

About a month later, in July 2025, Scale cut about 200 employees, roughly 14 percent of staff, and ended relationships with 500 global contractors. Droege told staff the company had scaled its core data-labeling business too quickly, and that it would hire for its enterprise and government sales units.²¹

New leadership and recent results#

On July 30, 2026 Scale announced that its board had appointed Francis deSouza as CEO, effective August 10, 2026, succeeding interim CEO Jason Droege. DeSouza was most recently chief operating officer and president of security products at Google Cloud, and earlier was CEO of Illumina.¹⁵

In a January 2026 post, Droege wrote that 2025 was Scale's strongest financial year, with well over $1 billion in new business, nearly half of it in the fourth quarter. He said the data business grew year over year and is now profitable, and that the applications business more than doubled revenue in the second half of 2025 and is expected to roughly double again in 2026. These are company statements, and the post gives no total revenue figure.¹⁴

Products and business model#

Scale's named products are the Scale data engine, the GenAI Portfolio and Donovan.³ The data engine page lists prompt-response generation, reinforcement learning from human feedback, red teaming and model evaluation, plus text, image, video and LiDAR annotation.⁴ On the applications side, Scale says it finds the use case, builds the system and takes ownership of the outcome for enterprises and governments.³ Its site also runs private leaderboards and benchmarks for AI companies, and in 2025 its research group SEAL says it introduced 15 new benchmarks and published more than 450 evaluations across more than 50 models.³¹⁴

Scale makes several marketing claims that we cannot verify: that 90 percent of the world's leading generative AI model builders are powered by it, that it has paid $1 billion to contributors worldwide, and that 15 billion human decisions have been used to train AI.³⁹ Treat these as company statements.

Data Partnerships program#

Scale's Data Partnerships page is the clearest public statement from either company about acquiring data from outside businesses. It asks established companies to share de-identified, structured records from everyday systems, including data warehouses such as Snowflake and BigQuery, chat tools, email and calendar, document stores, work-management tools and business platforms such as Salesforce and Zendesk. It describes the arrangement as a continuing partnership rather than a one-time transfer.¹⁶

The page makes these specific statements:

• Companies with about 40 or more employees are the preferred starting point, though each case is assessed individually.
• Partners are paid, and Scale gives an illustrative value of $10K to $1M-plus per partnership, scaling with cadence.
• Scale says it removes personally identifying information inside the customer's environment before extraction, and that partners approve each package before it leaves.
• Scale pays a referral bonus of up to $50,000 for introducing a company that signs a partnership.¹⁶

The page does not say which AI companies receive the data, and it does not address license terms, ownership of the data or exclusivity. Those would have to be negotiated.¹⁶ We do not know how many companies have signed, and the illustrative range is marketing, not a quote.

Government and enterprise work#

In March 2025 the Pentagon's Defense Innovation Unit selected Scale to lead Thunderforge, a prototype AI planning system for military operations with Anduril and Microsoft as partners, first for Indo-Pacific Command and European Command. The contract value was not disclosed.²² In February 2025 Qatar's government signed a five-year deal with Scale to deploy AI tools and training across public services, with more than 50 potential applications to be explored and no value disclosed.²³ Droege's January 2026 post said the Department of War awarded Scale two major contracts totaling nearly $200 million, and that Mayo Clinic, BP and Allianz joined as enterprise customers.¹⁴ In July 2026 Scale and Mayo Clinic described a collaboration on the Scale Generative AI Platform to cut record-review time and flag patient safety events, with patient data staying inside Mayo's environment.²⁴

Contributors: Outlier and Remotasks#

Scale operates Outlier, a platform where experts write prompts and correct answers, build grading rubrics, and rate AI responses. Its page says experts are in 50 countries, pay is weekly and tied to hours worked, there are no minimum hours, and more than $500 million has been paid to experts. It gives no specific rates.⁶

Earlier, Scale ran Remotasks. A 2023 Washington Post investigation, republished by Stars and Stripes, reported that of 36 current and former freelancers interviewed, all but two said payments had been delayed, reduced or canceled, and that workers in the Philippines often earned below the local minimum wage. Scale responded that its pay system is continually improving and that delays are exceedingly rare.⁷

Controversies and legal issues#

• Worker lawsuits (2024 to 2025). TechCrunch reported three suits in about a month: a December 2024 complaint, an early January 2025 suit by a former worker alleging she was paid below minimum wage and misclassified as a contractor, and a January 17, 2025 class action in the Northern District of California on behalf of six Outlier workers. The class action alleges they wrote disturbing prompts about violence and abuse without adequate psychological support and developed mental health problems. Scale said it complies with all laws, offers safeguards including opt-out and advance notice of sensitive content, and would defend itself.²⁵
• Labor Department probe (2025). The US Department of Labor reportedly dropped its Fair Labor Standards Act investigation of Scale in May 2025. TechCrunch said it was not clear why, and Scale said it was pleased with the update.²⁶
• Mercor suit (2025). On September 3, 2025 Scale sued rival Mercor for trade-secret misappropriation and former employee Eugene Ling for breach of contract, alleging he took more than 100 confidential documents. Mercor's co-founder said Mercor has no interest in Scale's trade secrets and had offered to have the files destroyed.²⁷

We found no reliable report of a final outcome in any of these matters. Allegations in lawsuits are not findings.

Invisible Technologies#

History and founding#

According to Forbes, founder Francis Pedraza raised $500,000 in 2015 to start the company, which combined remote workers with AI. After an early executive-assistant idea failed, it moved to back-office work. In March 2020 DoorDash asked it to help import restaurant menus and pricing, and about two years later OpenAI asked it to help reduce hallucinations in the model that became the basis for ChatGPT.⁸

Forbes reported in February 2025 that revenue grew from $3 million in 2020 to $134 million in 2024, with about $15 million of EBITDA, and that Pedraza had borrowed $20 million over three years to buy back shares from early investors after raising $23 million in total.⁸ Matthew Fitzpatrick, formerly global head of QuantumBlack Labs at McKinsey, joined as CEO in January 2025, and Pedraza is chairman and co-founder.¹⁰

Funding#

In September 2025 Invisible announced $100 million in growth funding led by Vanara Capital, an investment firm founded by TPG alumni, in its first publicly announced deal. New investors included Princeville Capital, HOF Capital, Freestyle VC, Rocketeer Management and Tallwoods Capital, and existing investors Acrew, Greycroft, Backed VC, BY Ventures and Deepwater took part. The company said total capital raised reached $144 million and did not disclose a valuation.¹⁰ Bloomberg reported at the time, citing people familiar, a valuation above $2 billion, and TechCrunch's list of 2025 US AI rounds shows $2 billion.¹¹²⁸ The company reports about 350 employees and offices in New York, San Francisco, Washington, D.C. and London, and says it opened a Warsaw office in 2025.¹⁰²⁹ We found no 2026 round.³⁰

Products and business model#

Invisible says it powers AI "from frontier model-building research to intelligent enterprise deployment" and claims to have trained models for more than 80 percent of the world's leading AI model providers. Its platform has five components:

• Neuron, a data platform that consolidates data from multiple systems into one governed place.
• Atomic, a process builder that documents workflows and routes steps between AI and people.
• Meridial, an expert network sourcing domain experts for projects.
• Synapse, an evaluation layer for quality, safety and accuracy.
• Axon, an agent platform that builds governed AI agents mirroring a customer's workflows.¹

For AI labs, Invisible lists annotation and evaluation, expert domain training, reinforcement learning environments that simulate business workflows, multimodal data work, red teaming, and training and evaluation in more than 80 languages. Its enterprise solutions include back-office automation, computer vision, contact center intelligence and demand forecasting.² Its case studies include Cohere, Boosted.ai, the Orchestra of St. Luke's, Alliance Nationwide Exposition and the Charlotte Hornets, and its September 2025 announcement names Microsoft, Swiss Gear, SAIC and the Hornets as users, along with a joint US Navy exercise with SAIC and Vatn Systems.¹¹⁰ Performance figures on those pages, such as an 80 percent average reduction in process complexity, are vendor statements.¹

Invisible says Deloitte ranked it 68th on the Technology Fast 500 with growth of 1,473 percent over the evaluation period, its second year on the list, and that the Inc. 5000 named it the second-fastest-growing AI company in 2024.³¹¹⁰

Contributors and Meridial#

Meridial is Invisible's expert network. Its page states that contributors are independent contractors, not employees, that every engagement is scoped by a statement of work showing the work, pay and project length, that rate ranges are published for each engagement and confirmed before work begins, and that access to the network does not guarantee projects. It cites 700-plus engagements across 12 domains.⁵ The page also uses the marketing phrase "millions of domain experts," while Invisible's home page claims 24,000-plus vetted experts; the two figures differ by orders of magnitude and we use the lower company figure.⁵¹

2026 developments#

On March 10, 2026 Invisible announced an agreement to acquire WeCP, an AI-native technical assessment platform founded in India that says it has run more than two million technical interviews and built more than 18,000 assessment frameworks. WeCP is to be folded into Meridial to expand expert evaluation and work on reinforcement learning environments. Terms and closing status were not stated.³² In January 2026 Invisible announced it had joined the World Economic Forum to contribute to research on AI in regulated environments.²⁹

Does Invisible buy data?#

Invisible's AI training page refers to a "Data" offering and to generating and labeling data, but it does not describe licensing or buying datasets from outside businesses.² Its enterprise products, Neuron and Atomic, are tools for the customer's own data and workflows.¹ That is a different model from Scale's Data Partnerships page.

Controversies#

We did not find reports of lawsuits or regulatory investigations against Invisible in the sources we opened. That may reflect a smaller company and less coverage, not a cleaner record, and we do not treat it as evidence either way.

Head to head#

Size and ownership#

Scale's press page cites a $29 billion valuation and more than 1,000 employees; Invisible reports a team of 350 and a reported valuation above $2 billion.⁹¹⁰¹¹ Neither publishes audited financials. Meta's 49 percent non-voting stake is the central ownership fact at Scale, while Invisible's largest recent round was led by a financial investor, Vanara Capital.¹⁸¹⁰

Business model#

Scale is a data and evaluation vendor that is expanding into applications for governments and enterprises, and says that applications revenue more than doubled in the second half of 2025.¹⁴ Invisible blends training data with enterprise data, process and agent software, aimed at a mix of labs and enterprise buyers.¹²

Neutrality and customer concentration#

Scale's relationship with Meta is the key independence issue. Droege said Scale remains independent, while OpenAI said it was winding down work with Scale and Google was reported to be planning to cut ties.¹⁹²⁰ Invisible has no strategic shareholder of that kind in the sources we read, and it lists logos spanning AWS, Microsoft and Cohere.¹ Neither company discloses customer concentration in the sources we reviewed.

Contributors#

Both rely on large independent-contractor pools. Scale has the longer and better documented history of disputes over pay and conditions, from Remotasks to the Outlier suits.⁷²⁵ Meridial's published terms are specific about statements of work and confirmed rates, while Outlier's page leads with total payouts and weekly pay.⁵⁶

Government work#

Scale's public-sector record is extensive, from Thunderforge to the Qatar deal and the Department of War contracts.²²²³¹⁴ Invisible lists public sector as a target industry and names SAIC as a customer of its platform, but names no government agency customers on the pages we read.¹¹⁰

Timeline of the public record#

• 2015: Invisible is founded; 2016: Scale AI is founded.⁸⁹
• March 2020: DoorDash asks Invisible to import restaurant menus and pricing.⁸
• 2022: OpenAI asks Invisible for help reducing hallucinations in its model.⁸
• May 2024: Scale raises $1 billion at a $13.8 billion valuation.¹²
• December 2024 to January 2025: Three worker lawsuits are filed against Scale; Fitzpatrick becomes Invisible's CEO in January.²⁵¹⁰
• February and March 2025: Qatar signs a five-year deal with Scale; the Defense Innovation Unit selects Scale for Thunderforge.²³²²
• May 2025: The Labor Department reportedly drops its probe of Scale.²⁶
• June 2025: Meta takes a 49 percent non-voting stake in Scale and Wang leaves.¹⁸¹³
• July 2025: Scale cuts about 200 employees and 500 contractors.²¹
• September 2025: Scale sues Mercor; Invisible announces its $100 million round.²⁷¹⁰
• January 2026: Droege calls 2025 Scale's strongest financial year; Invisible joins the World Economic Forum.¹⁴²⁹
• March 2026: Invisible agrees to acquire WeCP.³²
• July 2026: Scale and Mayo Clinic announce their collaboration on July 15; Scale announces deSouza as CEO on July 30.²⁴¹⁵
• August 10, 2026: DeSouza's tenure as CEO begins.¹⁵

How to read the numbers#

Three cautions apply to every figure in this comparison.

First, valuation is not business value. Scale's $13.8 billion and $29 billion figures reflect investor pricing at different dates and, in the Meta case, a non-voting stake sold at a price Scale did not itself state.¹²¹³ Invisible's "above $2 billion" is a Bloomberg report based on people familiar with the deal, and Invisible did not confirm it.¹¹¹⁰

Second, revenue is reported differently. Invisible gives a 2024 revenue figure of $134 million, which Forbes also reported, while Scale has given no revenue number in the materials we found and speaks instead of new business and growth rates.¹⁰⁸¹⁴ The two are therefore not directly comparable, and neither publishes margins beyond the profitability statements noted above.

Third, the contributor counts measure different things. Scale reports payouts ($1 billion to contributors globally, $500 million-plus through Outlier), Invisible reports a network size (24,000-plus experts), and Meridial's marketing refers to "millions."⁹⁶¹⁵

Strengths and weaknesses#

Invisible#

• Strengths: a profitable model by its own account and Forbes's EBITDA report, a broad stack beyond data labeling, and no strategic-investor conflict in the sources we read.⁸¹
• Weaknesses: much smaller scale, limited public detail on customers and terms, and no 2026 financing update.¹⁰³⁰

Scale AI#

• Strengths: the largest scale, deep government relationships, a long customer list, a new CEO with enterprise experience and a published data-partnership route for companies.^9141516
• Weaknesses: Meta's stake and reported customer departures, past layoffs, worker lawsuits and Remotasks criticism.^2021725

Who each suits#

Invisible suits enterprises that want training data and workflow automation from one vendor, and experts willing to take scoped project work through Meridial.¹⁵ Scale suits labs, large enterprises and public-sector bodies that value scale and an established vendor, experts who want Outlier's weekly-pay model with the caveat that pay depends on hours and project availability, and companies that want to explore a paid data partnership.³⁶¹⁶ AI labs that worry about a competitor's stake in a vendor may prefer vendors without one, a point the Scale case makes concrete.¹⁷

Questions to ask either vendor#

If you are evaluating either company as a supplier of data or evaluation services, or as a possible recipient of data, ask in writing:

1. Which of your engagements involve customer-supplied data, and how is it segregated from other clients' work?
2. Who among your contractors can access it, and under what confidentiality terms?
3. Whether any shareholder or strategic investor could access or benefit from the data, given the Meta precedent at Scale.¹⁸
4. Which AI companies receive the data or derived products, and whether you can limit that.
5. Who owns the data and any derived outputs, whether the arrangement is exclusive, and how long data is retained after termination.
6. What certifications and incident history you can document.

These are generic diligence questions. The public pages we reviewed do not answer them for either company, and Scale's Data Partnerships page explicitly leaves license, ownership and exclusivity terms unaddressed.¹³¹⁶

What this means for data sellers#

For a business thinking about selling or licensing its data, the pair gives two different lessons. Scale publishes an invitation to share de-identified operational records for payment, with an illustrative value range and a company-size guideline, which makes it a reachable route for qualifying companies.¹⁶ Invisible's public materials describe expertise and software delivered as a service, and we found no data-acquisition program.²¹

Four practical points follow:

1. Read the terms, not the range. Scale's $10K to $1M-plus is described as illustrative. The page does not say whether the arrangement is a sale, a license or a revenue share.¹⁶
2. Ask who the end buyers are. Scale's page refers only to "AI companies," and the disputes and ownership questions around Scale show why a licensor should ask who the ultimate recipients are and whether ownership or contracts could change.¹⁶¹⁹
3. De-identification is a claim to test. Scale says it anonymizes inside your environment. Ask how, who audits it, and what happens if re-identification is possible.¹⁶
4. Understand the contractor side separately. Both recruit individual experts. If your staff have specialist expertise, that is a labor arrangement with its own classification and pay terms, not a data license.⁵⁶

Our guide to selling data to AI companies, the buyers page and the trust page cover what to prepare. SellYourAIData is not affiliated with Invisible or Scale AI, has no partnership with either, and nothing here suggests either will buy your data.

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.

What we could not verify#

• Scale's revenue. We found no first-party or news figure we could open for 2024 or 2025, so we do not state one; Scale's own post gives growth statements only.¹⁴
• Whether the $14.3 billion Meta figure matches the final amount; Scale did not state it, and the figure rests on Reuters reporting from two sources.¹³
• Outcomes of the Scale worker suits and the Mercor case; we found no reliable report of a resolution.²⁵²⁷
• Scale's customer list after 2025, beyond what its own site and blog name.³¹⁴
• Invisible's contractor pay rates and any disputes; its published pages give no rates.⁵
• How many companies have signed Scale data partnerships, and on what terms.¹⁶
• Neither company's 2026 revenue is public in the sources we found.

Related reading#

See also Scale AI vs Surge AI and Scale AI vs Labelbox.