Mercor vs Scale AI: two different businesses with the same customers
Mercor and Scale AI appear on the same shortlists and get compared as though they are alternatives. They are not. Scale sells a managed data engine and, increasingly, government software; Mercor runs a marketplace that sources credentialed experts and bills their hours at a markup. The two do overlap at the point of purchase, because a frontier lab needing expert reasoning data can buy it from either, but the delivery model, the accountability, and the risks you take on differ substantially. This comparison works through what each actually sells, what each publishes about itself, and the conditions under which each is the right choice.
Key takeaways
- Scale delivers finished work under contract. Mercor sources and matches people, bills their time at a markup, and increasingly sells finished assets too. Confirm which you are buying.
- Scale carries a neutrality question that Mercor does not. Meta owns 49 percent of Scale, and several of its largest customers pulled back within weeks of the 2025 investment.
- Mercor's widely cited revenue figure is gross marketplace volume, not net revenue. Most of it flows to contractors, and Mercor does not publish the net.
- Scale holds the strongest government compliance posture in the category. Mercor's trust center lists SOC 2 Type 2 and nothing else.
- Neither publishes IP indemnification terms. Mercor's public terms of service run the indemnity in the opposite direction and cap liability at 100 dollars.
- Neither publishes a named enterprise case study with attributable metrics.
About Mercor
Mercor was founded in January 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha, and is based in San Francisco. It raised a 100 million dollar Series B at a 2 billion dollar valuation in February 2025, then a 350 million dollar Series C led by Felicis at a 10 billion dollar valuation in October 2025. A further round at roughly 20 billion was reported in talks from July 2026, with Nvidia reported as a potential investor in August 2026. As of this writing there is no report that it has closed, so treat the 20 billion figure as a negotiation, not a valuation.
The model is an expert marketplace. Candidates complete an AI-run interview of roughly twenty minutes, which reviews their background and probes technical depth, with up to three retakes permitted. Qualified experts are matched to client projects, and Mercor bills the client a marked-up hourly rate. Mercor states an average contracted rate of 107 dollars per hour, with sample listings ranging from 60 dollars per hour for language tutors to 450 dollars per hour for oncologists.
Mercor also runs a robotics line that is genuine physical collection: it supplies capture kits to partner businesses, employees opt in and wear cameras during shifts, and Mercor licenses the resulting footage to AI and robotics companies, paying 12 to 18 dollars per approved hour.
Named benchmark partners on its own site include Harvey, Ramp, Cognition, and Sierra. Third-party reporting names OpenAI, Anthropic, Meta, and Google as customers, and Nvidia has disclosed using Mercor data in recent Nemotron models.
About Scale AI
Scale AI was founded around 2016 by Alexandr Wang and Lucy Guo, and is based in San Francisco. In June 2025, Meta invested roughly 14.3 billion dollars for a 49 percent stake at a valuation above 29 billion. Wang left to lead Meta's superintelligence team while remaining on Scale's board, and Francis deSouza became chief executive in August 2026.
Scale sells a Data Engine covering annotation and curation across text, image, video, and 3D and LiDAR; a generative AI portfolio covering RLHF, prompt-response generation, and red-teaming; public model leaderboards; and Donovan, a decision-support platform sold into defense and enterprise. Its public positioning has shifted substantially toward government and enterprise software.
Its government business is the most verifiable part of its record, because federal contract data is public. Army contract W911QX20C0051 shows 105.5 million dollars total obligated across modifications; an Army indefinite-delivery contract signed in May 2026 carries a ceiling near 99.5 million; and a February 2026 Air Force award of roughly 5.1 million covers integrating Donovan into an airborne command platform. The widely cited 500 million dollar Department of Defense agreement is a ceiling rather than obligated spend, which is a distinction worth preserving when the number appears in vendor marketing.
Named customers with case-study-level description are Mayo Clinic and British Petroleum. Scale claims that 90 percent of leading generative AI model builders are powered by Scale, stated without methodology.
What buyers report
Neither company has a usable enterprise review base, which is itself the finding.
Scale's GenAI Platform holds 4.5 out of 5 on Gartner Peer Insights from two reviews. Reviewers cite model quality and the combination of models, infrastructure, and tooling, and flag pricing structure as a purchasing concern. With a sample of two, this is an anecdote rather than a signal.
Mercor holds 4.4 out of 5 on G2 across 114 reviews, which looks substantial until you check what is being reviewed. Mercor's G2 profile sits under applicant tracking, AI recruiting, and job search categories, so the reviews come from candidates and contractors rather than from enterprises buying data. Mercor has no Gartner Peer Insights page at all. Its Trustpilot reviews are contractors and its Glassdoor reviews are employees. Every public rating measures the labor-supply side.
For both companies, the practical consequence is that a paid pilot on your own task will tell you more than any available third-party source.
The comparison
| Mercor | Scale AI | |
| Model | Expert marketplace, hours billed at markup, plus some finished assets | Managed data engine and government software platform |
| Accountability | You typically define the task and manage quality | Scale contracts for delivery |
| Modalities | Text and code primarily; robotics video via partner businesses | Text, image, video, 3D and LiDAR, multimodal |
| Workforce disclosure | 107 dollars per hour average rate published; 107 payout countries published; network size self-contradictory | No contributor headcount, geography, or pay rates published |
| Compliance | SOC 2 Type 2 only | SOC 2 Type II, ISO 27001, DoD IL4, FedRAMP High; no HIPAA found |
| Government | Not a government vendor | Strongest posture in the category, with verifiable contract history |
| Open data | None found | 36 datasets and 5 models on Hugging Face, the largest open footprint of any services vendor |
| Pricing | No rate card; marked-up hourly | Not published |
| IP indemnification | Not published; public terms indemnify Mercor, not you | Not published |
Where the neutrality question lands
Scale carries a trust problem that Mercor does not, and it is structural. Meta's 49 percent stake means a frontier lab routing unreleased-model data through Scale is routing it through a pipeline part-owned by a competitor. Within weeks of the June 2025 investment, Google, OpenAI, and xAI were reported to have pulled back or halted projects. By August 2025, reporting indicated Meta's own AI unit was favoring other vendors, with internal sources describing Scale's data as low quality, which Meta disputed. The ownership is permanent until it changes; the concern does not resolve with time. For a lab with competitive exposure to Meta this is the single largest factor in the comparison, and no amount of contractual language resolves it, because the issue is who owns the counterparty rather than what the counterparty has promised.
Reading Mercor's revenue figure correctly
Mercor is frequently described as a 2 billion dollar revenue company. The Information reported that figure in June 2026 and labeled it gross annualized revenue, which on a marketplace means total customer spend before contractor payouts. Contractor share is estimated at 60 to 70 percent, so net revenue is a fraction of the headline. Mercor does not publish net revenue, and third-party estimates span roughly 180 million to 800 million dollars, which is a range wide enough to be uninformative. Mercor's own chief executive gave growth timeframes in a single Fortune interview that conflict with each other, which the publication noted without reconciling.
This matters for a buyer in one narrow way: vendor scale is often used as a proxy for stability, and a gross marketplace figure is a poor proxy. Use the funding record, which is documented, rather than the revenue claim.
Contract terms, where the gap is widest
Scale's public pages carry no IP indemnification commitment, which is not evidence of its absence in a negotiated enterprise contract but does mean you have to ask.
Mercor's published terms of service are more specific and run the other way. The only indemnity flows from the user to Mercor, covering violations of third-party rights including intellectual property. The terms state that users do not own their account or possess rights to data stored by Mercor, that all usage data is owned exclusively by Mercor, and that ownership of user content may be governed by separate agreements. Liability is capped at the greater of the amount paid or 100 dollars. These are platform and contributor-facing terms rather than an enterprise agreement, and no public enterprise agreement exists, so the terms you would actually sign are unknown from outside.
For any program where training-data provenance could later be litigated, get both ownership and indemnification in writing before scoping.
Which to choose
Choose Scale when you need breadth across modalities including 3D and LiDAR, when you want a single vendor contractually accountable for delivery, and above all when you are deploying into a defense or government environment, where FedRAMP High and DoD IL4 effectively decide the question. Weigh the Meta ownership seriously if you are a lab with competitive exposure.
Choose Mercor when what you lack is access to credentialed people rather than the capacity to manage them, when the work is text and reasoning in professional domains, and when speed to staff matters more than contractual accountability for the output. Ask about security posture since the 2026 incident, and ask what certifications beyond SOC 2 Type 2 now exist.
Why HumanSignal is the better choice when the data does not exist yet
Both companies in this comparison answer the same question: who does the work. Mercor answers it with a marketplace of credentialed people billed by the hour, Scale with a managed engine that labels what you send it. Neither answers the question underneath it, which is whether the examples your model needs have been recorded at all. If they have not, a marketplace will sell you qualified people with no protocol to follow, and a data engine will wait for data that is never going to arrive.
HumanSignal Services is built for that case specifically. We design the collection protocol, recruit for the physical or professional characteristics the protocol requires, run the capture in our studios, on site at your facility, or in the field with synchronized motion, vision, audio, and sensor streams, and deliver a dataset verified against the spec we agreed. The accountability sits with us rather than with a staffing layer, which is the practical difference between commissioning a dataset and hiring people to attempt one.
The quality mechanism is visible rather than asserted. Review workflows run in Label Studio Enterprise, the commercial edition of the open source labeling tool we build, so reviewer decisions, inter-annotator agreement, and rework rates are reportable to you during the engagement instead of summarized afterward. Three further things we publish that neither company here does: contributor pay ranges, so our sourcing claims are checkable; a customer agreement giving you exclusive ownership of customer data and of insights produced specifically for you; and SOC 2 Type 2 and HIPAA compliance documented on our security page.
The capability came from acquiring Erud AI, a physical data collection company, in November 2025, and it is now the core of what HumanSignal Services does. If your constraint is capture rather than staffing, the physical AI data services page sets out how a program gets scoped, and we will walk the protocol and the acceptance criteria with you before anything is committed.
Is Mercor a data labeling company
Not in the traditional sense, and it does not describe itself that way. Mercor sources and matches credentialed experts who produce post-training data, evaluation, and rubrics, billing their hours at a markup. It also licenses wearable-camera footage from partner businesses and sells some off-the-shelf data. If you need managed annotation of an existing corpus, it is not the closest fit.
Does Meta's stake in Scale AI actually affect customers
It did measurably. Within weeks of the June 2025 investment, Google, OpenAI, and xAI were reported to have pulled back or halted projects, and Scale laid off roughly 200 data-labeling staff. Whether it affects you depends on whether Meta is a competitor in your market. For enterprises outside frontier model development, it is usually a smaller consideration than compliance scope.
Which one is cheaper
Neither publishes a rate card and the models are not comparable per unit. Mercor's cost is a marked-up hourly rate that varies by domain, from around 60 dollars per hour to above 400 for clinical specialists. Scale prices per engagement. Compare total program cost including your own management overhead, which is substantially higher on a marketplace model because task definition and quality control stay with you.