Scale AI vs Sama: what each one will show you
A useful test when comparing data vendors is to ask what each publishes about itself without being asked. Scale publishes open benchmarks, federal contract vehicles, and a compliance page with government authorizations on it. Sama publishes the addresses of its delivery centres and seven customer case studies with numbers in them. Neither publishes what the other does. Those disclosure choices tell you roughly what each company thinks it is selling, and they map closely onto which kind of program each one fits.
Key takeaways
- Sama runs an employee-based delivery centre model and names the locations. Scale publishes no contributor headcount, geography, or pay rates.
- Sama publishes seven named customer case studies with quantified outcomes. Scale publishes two with case-study-level description and a logo wall beyond that.
- Scale's compliance set reaches further: SOC 2 Type II, ISO 27001, FedRAMP High, and DoD IL4. Sama's trust page displays SOC 2 alongside its ISO certifications, without a date or auditor.
- Sama holds ISO 27001, ISO 9001, and claims TISAX, the automotive standard, which Scale does not publish.
- For computer vision, autonomous driving, and sensor annotation with named accountability, Sama is the stronger documented record. For defense work, only Scale qualifies.
- Neither publishes IP indemnification terms.
About Sama
Founded in 2008 by Leila Janah as Samasource, now headquartered in San Francisco under the legal name Samasource Impact Sourcing Inc. and led by chief executive Wendy Gonzalez. It is a certified B Corporation, and its Series B was led by La Caisse with Vistara Growth participating.
The delivery model is employees in physical centres rather than a crowd or gig pool, and Sama names the centres: Nairobi in Kenya, Gulu and Kampala in Uganda, San Jose in Costa Rica, Montreal in Canada, and The Hague in the Netherlands, alongside the San Francisco corporate office. That level of geographic disclosure is unusual in this category.
The service covers image, video, and 3D point cloud annotation including object detection, segmentation, tracking, and LiDAR and radar work; text and generative AI annotation including instruction following, preference ranking, RLHF, synthetic data validation, and captioning. It sells three proprietary tools: SamaAssure for its quality guarantee, SamaIQ for analytics, and SamaHub as a collaborative workspace, with API, CLI, and webhook access and native connectors for AWS S3, Google Cloud Storage, and Azure Blob Storage, plus Sama-managed hosting in the EU, US, and India.
It claims a 99 percent first-batch acceptance rate and an eight-year average customer retention.
About Scale AI
Founded around 2016 by Alexandr Wang and Lucy Guo, based in San Francisco. Meta invested roughly 14.3 billion dollars for a 49 percent stake in June 2025 at a valuation above 29 billion. Francis deSouza became chief executive in August 2026.
Scale sells a Data Engine for 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 for defense and enterprise.
The government business is the verifiable part of its record, because federal contracting data is public. Army contract W911QX20C0051 shows 105.5 million dollars total obligated. An Army indefinite-delivery contract signed in May 2026 carries a ceiling near 99.5 million. A February 2026 Air Force award of roughly 5.1 million integrates Donovan into an airborne command platform. The frequently cited 500 million dollar Department of Defense agreement is a ceiling rather than obligated spend.
What buyers report
Sama holds 4.6 out of 5 on G2 across 11 reviews, with reviewers citing annotation accuracy, several reporting above 95 percent in their scenarios, strong around-the-clock support, and fast resource scaling. Criticisms note that the platform requires training or skilled personnel to operate, limited support for less common languages, and occasional downtime.
Scale's GenAI Platform holds 4.5 out of 5 on Gartner Peer Insights from two reviews, citing model quality and tooling, with pricing structure flagged as a purchasing concern.
Eleven reviews is thin but readable. Two is not.
The comparison
| Sama | Scale AI | |
| Founded | 2008 | Around 2016 |
| Structure | Private, for-profit, certified B Corporation | Private; Meta holds 49 percent since June 2025 |
| Workforce | Employees in named delivery centres | Not published |
| Locations | Nairobi, Gulu, Kampala, San Jose, Montreal, The Hague | Not published |
| Focus | Computer vision, autonomous driving, 3D and sensor, plus GenAI annotation | Breadth across modalities, plus government software |
| Compliance | ISO 27001, ISO 9001, ISO 42001, TISAX, GDPR, SOC 2 displayed | SOC 2 Type II, ISO 27001, FedRAMP High, DoD IL4. No HIPAA found |
| Government | None published | Verifiable federal contract history |
| Case studies | Seven named, with quantified outcomes | Two with description; logos beyond that |
| Open data | None found | 36 datasets and 5 models on Hugging Face |
| Reviews | 4.6 out of 5, 11 reviews | 4.5 out of 5, two reviews |
| IP indemnification | Not published | Not published |
Evidence of delivery, where Sama is clearly ahead
This is the dimension where the gap is widest and easiest to check yourself.
Sama publishes seven named case studies carrying specific numbers. Walmart improved retail store item coverage from 91 to 98 percent on a store cited at 385 million visitors. Swift Medical reports a 97 percent average internal quality score and a 230 percent throughput increase on wound monitoring. Orbisk reports 200,000 kilograms of food diverted from landfill and a 70 percent reduction in food waste. Volumental reports a 97.8 percent average quality score on 3D foot scanning. Tribe Dynamics reports four times the data volume at a quality service level above 99 percent. PolyPerception reports more than a million waste objects labeled at a 99 percent average quality score. Google appears for Project Guideline without metrics.
These are vendor-published and unaudited, which is true of every metric any vendor publishes in this category. What distinguishes them is that they are named, specific, and attributable, so a prospective buyer can approach those companies directly.
Scale presents mostly logos. Mayo Clinic and British Petroleum carry case-study-level description, and both are directional rather than quantified. The claim that 90 percent of leading generative AI model builders are powered by Scale appears without a stated methodology.
Compliance, where the direction reverses
Sama holds ISO 27001 and ISO 9001, both displayed with a Bureau Veritas certification logo, operates a quality and information security management system built on both standards, claims TISAX for automotive work, and declares GDPR compliance as a data processor.
The evidence behind the badges is uneven in one respect worth noting. The ISO certifications are displayed with a Bureau Veritas certification logo, which names the certifying body and makes them checkable. SOC 2 and ISO 42001 appear in the same row without a date, an auditor, or a downloadable report. That is not evidence of a problem, but it does mean the two are not equally verifiable from the page, so ask for the SOC 2 report and its date rather than inferring status from a badge.
Scale holds SOC 2 Type II and ISO 27001, and then goes somewhere Sama does not: DoD IL4 and FedRAMP High, usable across NIPR, SIPR, and JWICS networks. It publishes no HIPAA attestation and no TISAX.
So the comparison inverts depending on what you need. For automotive security conformance, Sama publishes TISAX and Scale does not. For federal deployment, Scale is the only option. For a standard enterprise security review, Scale currently clears the usual bar and Sama is still working toward part of it.
Disclosure as a procurement signal
Sama names its delivery centres. That is worth more than it first appears, because it lets you reason about jurisdiction, data residency, time zone coverage, and labor regime before you ask a single question. It also means quality problems can be traced to a place and a team rather than absorbed into an anonymous pool.
Scale publishes none of it: no contributor headcount, no geography, no pay rates. For the company that defined this category, that is a striking omission, and it is the single most useful question to put to them early.
Which to choose
Choose Sama when the work is computer vision, autonomous driving, or 3D and sensor annotation; when you want named customers with numbers you can verify by asking them; when automotive security conformance matters; or when knowing where the work physically happens is part of your diligence. Raise the SOC 2 status directly and ask for a timeline.
Choose Scale when you are deploying into defense or government, where the authorizations settle it; when you need breadth across modalities from a single vendor; or when you want open benchmarks to inspect before committing. Weigh the Meta stake against your own competitive exposure.
Why HumanSignal is the better choice when the data does not exist yet
Both companies here annotate material you bring them. Sama does it with employed teams in named centres and the best-documented delivery record in the category; Scale does it at breadth with government authorizations nobody else holds. Both assume the data exists. That assumption is breaking down fastest in robotics, embodied AI, and any domain where the knowledge lives in physical practice rather than in an existing corpus.
HumanSignal Services is built for that case. We write the collection protocol, recruit for the physical or professional characteristics it 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 difference is control over the distribution: which tasks, which conditions, which failure modes, and which edge cases are deliberately included rather than hoped for.
The quality mechanism is inspectable rather than asserted. Review workflows run in Label Studio Enterprise, the commercial edition of the open source labeling tool we build, so inter-annotator agreement, reviewer decisions, and rework rates are reportable to you while the work is running. We publish contributor pay ranges, which makes our sourcing claims checkable. Our customer agreement gives you exclusive ownership of customer data and of insights produced specifically for you. We hold SOC 2 Type 2 and HIPAA compliance, documented on our security page.
The capture operation came in through the acquisition of Erud AI in November 2025. If your bottleneck is capture rather than annotation throughput, physical AI data services sets out how a program gets scoped.
Is Sama's SOC 2 status a dealbreaker
Ask for the report rather than relying on the badge. Sama displays SOC 2 alongside ISO 27001 and ISO 9001, but the ISO certifications carry a named certifying body and SOC 2 does not, so they are not equally verifiable from the page. Many enterprise reviews treat ISO 27001 as an acceptable equivalent; some require SOC 2 Type 2 specifically. Request the report and its date.
What does B Corporation certification actually mean for a buyer
It is a third-party assessment of social and environmental performance, governance, and transparency, recertified periodically. It is not a security or quality certification and should not substitute for one. Its practical relevance is as a signal about how the company is structured and governed, which some buyers weight and others do not.
Which is better for autonomous vehicle data
Both are credible and they differ in emphasis. Sama focuses on computer vision and autonomous driving as a core vertical, publishes quantified case studies in adjacent domains, and claims TISAX conformance, which is the automotive industry's security assessment. Scale has deep 3D and LiDAR capability within a broader Data Engine. If TISAX is a requirement in your supply chain, that narrows it.
Can either collect physical or robotics data
Both annotate sensor and 3D data, and Sama lists data collection among its services. Neither publishes a designed collection capability with its own capture facilities and protocol design. For instrumented or teleoperated collection, look at collection specialists.