Meta Description: Compare the top Scale AI alternatives for 2027, from RLHF and expert data vendors to computer vision annotation firms, with headquarters, services and best fit.
In June 2025, Meta invested about $14.3 billion in Scale AI for a 49% stake and hired founder Alexandr Wang to lead a new AI unit, according to Fortune. The reaction from other model builders was quick. OpenAI said it was phasing out its work with Scale, and Reuters reported that Google planned to cut ties. A month later, interim CEO Jason Droege laid off about 200 full-time employees, roughly 14% of staff, citing shifts in market demand, as TechRepublic reported. In July 2026 Scale named former Google Cloud executive Francis deSouza as CEO, effective August 10, 2026.
Scale is still operating and still signing customers. Reuters noted that it recently added BP and Mayo Clinic as enterprise clients. But for AI labs that compete with Meta, and for enterprises that would rather not route proprietary data through a company 49% owned by one, the question has shifted from “should we use Scale?” to “who else can do this work?” The vendor market has answered with expert networks for post-training, managed workforces for computer vision, and platforms that let teams run labeling in-house.
This list covers ten Scale AI alternatives across both halves of what Scale sells: generative AI data (RLHF, supervised fine-tuning, evaluations, red teaming and RL environments) and classic annotation (image, video, LiDAR and sensor fusion). Every company is independent of Meta and was operating as of October 2026. For a broader view of the market, see our ranking of the top data annotation and labeling companies.
Scale AI Alternatives At A Glance
Company
Headquarters
Best For
Core Services
Surge AI
San Francisco, California
Frontier labs replacing Scale’s RLHF and post-training work
Post-training runs, expert workforce, SWE data, benchmarks
Mercor
San Francisco, California
Labs that need credentialed domain experts at volume
Expert sourcing, human data and evals, RL environments, benchmarks
Turing
San Francisco, California
Coding and reasoning data for LLM post-training
Coding data, evals, preference pairs, RL, enterprise AI builds
Snorkel AI
Redwood City, California
Enterprises building specialized models and evaluations
Teams that want a platform plus an on-demand expert network
Labeling platform, Alignerr experts, RL environments, robotics data
Invisible Technologies
New York, New York
Enterprises pairing model training with workflow automation
AI training, RL environments, expert network, evaluation, agents
iMerit
San Jose, California
Autonomous vehicle, medical imaging and robotics annotation
LiDAR and sensor fusion, DICOM, Ango Hub, RLHF via Scholars
TELUS Digital
Vancouver, Canada
Multilingual and automotive data at enterprise scale
2D/3D sensor fusion, data collection, post-training, red teaming
Appen
Sydney, Australia
Global language coverage and long-running data programs
Annotation, speech data, SFT, RLHF, red teaming, ADAP platform
Sama
San Francisco, California
Computer vision teams that prioritize first-pass quality
Image, video and 3D point cloud annotation, model evaluation
Best Scale AI Alternatives In 2027
1. Surge AI — Bootstrapped RLHF Specialist With More Than $1 Billion In Revenue
Surge AI is the company most often named as Scale’s direct replacement for frontier post-training work. Founded in 2020 by Edwin Chen, a former Google and Meta engineer, and based in San Francisco, it grew without outside funding to more than $1 billion in annual revenue while staying profitable, according to SiliconANGLE’s summary of Reuters reporting from July 2025. The same coverage reported that OpenAI and Google, both of which moved work away from Scale, along with Anthropic, were Surge customers. Reports at the time said Surge was seeking its first outside capital at a valuation of $15 billion or more.
What a Scale customer would compare is the post-training stack. Surge sells post-training runs and off-the-shelf datasets, staffs projects from an expert workforce and a consultant network, and runs a dedicated software engineering track for coding data. It also publishes its own benchmarks, including Riemann-bench, Hemingway-bench and the DAYJOB series for tax, healthcare and finance tasks, which signals the kind of expert evaluation work it takes on.
Surge is a narrow fit by design. It is built for labs and large model teams buying human preference data, demonstrations and evaluations, not for companies that need bounding boxes on dashcam footage. Buyers coming from Scale’s GenAI business will find the closest like-for-like here. Buyers coming from Scale’s autonomous vehicle or mapping work should look further down this list.
2. Mercor — Expert Network Paying An Average Of $93 Per Hour
Mercor, founded in 2023 and headquartered in San Francisco, began as an AI hiring platform and now connects AI labs with domain experts such as physicians, lawyers, engineers and consultants. Its homepage lists more than 5 million experts, an average contracted rate of $93 per hour and more than $4 million in daily payouts. Mercor raised a $350 million Series C at a $10 billion valuation in October 2025, led by Felicis Ventures, according to TechCrunch.
For a Scale customer, Mercor is the option when the bottleneck is finding qualified people rather than managing a labeling pipeline. It supplies human data and evaluations, builds RL environments, and runs the APEX benchmarks with partners such as Harvey, Ramp and Cognition. In July 2026, CEO Brendan Foody said gross annualized revenue had passed $2 billion, and Mercor announced it was buying Deeptune, which builds simulated environments for AI agents, as The Next Web reported.
Buyers should factor in security history. In March 2026, Mercor disclosed a breach linked to a compromised release of the open-source LiteLLM library. Meta paused its work with the company, contractors filed lawsuits, and TechCrunch reported that other model makers were reviewing their exposure. According to The Next Web, OpenAI and Anthropic remained customers. Ask Mercor for its current security attestations and incident remediation details before sending sensitive training protocols.
3. Turing — Coding And Reasoning Data From A 4M+ Talent Cloud
Turing started in 2018 as a platform for sourcing remote software engineers and turned that talent pool into a training data business. It is headquartered in San Francisco and led by CEO Jonathan Siddharth. TechCrunch reported in March 2025 that OpenAI had asked Turing to generate code to improve its models’ reasoning, and that Turing raised $111 million in a Series E at a $2.2 billion valuation, with annualized revenue of about $300 million at the time.
The overlap with Scale is strongest in coding and STEM post-training. Turing says it works with leading AI labs on reasoning, coding and multimodality, drawing on a talent cloud of more than 4 million software engineers, data scientists and STEM experts. Its ALAN platform matches and manages that talent and generates both human and synthetic data for evals, fine-tuning, reinforcement learning, preference-pair generation and benchmarking.
Turing also runs Turing Intelligence, which builds AI systems for enterprises by fine-tuning models on their own data. That gives enterprise buyers a single vendor for both training data and deployment. Scale has been building a similar split between lab work and enterprise work. Teams whose Scale spend was mostly code generation, code review or math reasoning data should put Turing on the shortlist, as should any team that needs software engineers who can work inside real repositories rather than generalist raters.
4. Snorkel AI — Stanford AI Lab Spinout Valued At $3.5 Billion
Snorkel AI was founded in 2019 out of the Stanford AI Lab and is based in Redwood City, California, with co-founder Alex Ratner as CEO. It started with programmatic data labeling software and has moved toward expert data as a service. In September 2026 Snorkel raised a $350 million Series E at a $3.5 billion valuation, led by Insight Partners and S32, roughly triple the $1.3 billion valuation of its May 2025 Series D.
Snorkel’s current offering maps closely to Scale’s evaluation and data products. Snorkel Expert Data-as-a-Service delivers expert-curated datasets, demonstrations, preference labels, rubrics and standard or custom environments. Snorkel Evaluate is used to build specialized evaluations for models and agents. The company also publishes leaderboards across coding, computer use, finance and legal tasks, and has committed $30 million to an Open Benchmarks Grants program.
The customer base spans both labs and regulated enterprises. Snorkel’s Series D announcement named BNY, Wayfair, Chubb and the U.S. Air Force, and its homepage lists Anthropic, OpenAI, Google and Mistral AI among the organizations it works with. That mix makes Snorkel a sensible alternative for financial services, insurance and government buyers that want domain experts and evaluation rigor without a crowd marketplace. Its background in programmatic labeling also suits teams that want to combine expert labels with weak supervision on large internal datasets.
5. Labelbox — Labeling Platform Plus 2.6M+ Skill-Tested Alignerr Experts
Labelbox, founded in 2018 and headquartered in San Francisco’s Mission District, is the alternative for teams that want to own the tooling. It began as a data labeling platform for images, video and text and has since added services on top. Its Alignerr network, launched in November 2024, gives buyers access to vetted experts who rate, evaluate and label generative AI data, and Labelbox now cites more than 2.6 million skill-tested experts across 200+ domains.
The newer products line up with Scale’s frontier lab work. Horizon provides RL environments and evals for reasoning, tool use and computer use. Terra captures robotics training data from purpose-built rigs with seven synchronized viewpoints. Labelbox helped Meta build the GIM benchmark of 820 expert-authored problems and says more than 90% of leading U.S. AI labs post-train and evaluate on its platform. In October 2026 it closed a Series E of more than $100 million, combining equity and credit, led by Infinitum Partners with Wipro Ventures, Andreessen Horowitz and B Capital participating.
Labelbox works in two modes. Teams can run their own annotators on the platform, or bring in Alignerr experts and plug them into existing pipelines, as Ideogram did to build an expert labeling team with its own screening criteria. Named customers include Walmart, Pinterest, Etsy, Stryker and Intuitive Surgical. For teams comparing tools rather than vendors, our list of AI data annotation platforms covers more options.
6. Invisible Technologies — RL Environments And 24,000+ Vetted Experts For Enterprise AI
Invisible Technologies, founded in 2015 and based in New York, describes itself as an applied AI research company. It began as a virtual assistant service and grew into AI training work, with RLHF contracts for Microsoft and Cohere as its main revenue source, according to Sacra. Former McKinsey executive Matthew Fitzpatrick became CEO in 2024. In September 2025, Bloomberg reported that Invisible raised $100 million at a valuation of more than $2 billion, led by Vanara Capital, and described it as a Scale AI rival.
On the lab side, Invisible supplies training data and expert-built RL environments for AI agents and says it has trained more than 80% of the world’s leading AI models. On the enterprise side, it sells a five-part platform: Neuron for preparing multi-source data, Atomic for mapping and automating workflows, Meridial for its network of more than 24,000 vetted domain experts, Synapse for evaluating accuracy and safety, and Axon for building governed agents.
That enterprise layer is the main difference from a pure data vendor. Invisible works across banking, insurance, healthcare, life sciences, private equity and the public sector, and lists computer vision, contact center and back-office automation among its solutions. Named clients include Cohere, Microsoft, AWS, ElevenLabs and Thomson Reuters. It suits buyers that used Scale for both model data and enterprise AI projects and want one partner for each, rather than a vendor that only ships labeled datasets.
7. iMerit — LiDAR And Medical Annotation Under SOC 2, ISO 27001 And TISAX
iMerit is one of the strongest classic annotation alternatives on this list. Founded in 2012 by Radha Basu and based in San Jose, California, it built its business on large in-house teams labeling images, video and sensor data. In August 2026, EXL completed its acquisition of iMerit in a deal worth up to $310 million, including $170 million upfront, according to EXL’s filing with the SEC. iMerit now operates as part of EXL.
For buyers replacing Scale’s autonomous vehicle and mapping work, iMerit covers image, video, LiDAR, 3D point cloud and sensor fusion labeling, plus HD mapping, 3D segmentation and support for vision-language-action models. Its healthcare practice handles medical imaging, digital pathology, DICOM annotation, clinical notes and de-identification, which is useful for buyers who also need the specialists in our medical data annotation companies ranking. The company lists SOC 2, ISO 27001, GDPR, HIPAA and TISAX compliance. TISAX matters to European automakers.
iMerit also competes for GenAI work. Its Scholars network of more than 25,000 domain experts in 60+ countries handles RLHF, model evaluation, chain-of-thought reasoning and red teaming, and its Ango Hub platform supports annotation, QA and evaluation across image, video, text, audio and DICOM data. Clients shown on its site include Johnson & Johnson, GE Healthcare, Blue River Technology, Carbon Robotics and Autodesk.
8. TELUS Digital — 500+ Annotation Languages And 2D/3D Sensor Fusion Labeling
TELUS Digital, headquartered in Vancouver, was known as TELUS International until its 2024 rebrand. Its AI data business grew through acquisitions, including Lionbridge AI, which it agreed to buy for $935 million in late 2020, and Bengaluru annotation firm Playment in 2021. In October 2025, parent company TELUS completed a take-private deal for the shares it did not already own, paying about $539 million. TELUS Digital is now wholly owned by TELUS.
For a Scale customer, TELUS Digital’s appeal is breadth. It reports a global AI community of more than 1 million contributors, more than 500 annotation languages and dialects, and more than two billion labels delivered each year. Its automotive practice covers in-cabin and open-road data collection, 2D and 3D sensor fusion annotation, and training data for CNN and vision-language models, with collection operations in more than 50 countries. It also offers off-the-shelf datasets for LLMs, speech recognition and automotive models.
On the GenAI side, TELUS Digital sells expert post-training data, chain-of-thought reasoning, preference tuning, red teaming and safety evaluations across more than 20 domains, as well as physical AI data such as egocentric video collection. Search and ads rating is another long-running line of work. The company says it serves more than 600 clients across its business. It fits multinational buyers that need many languages and regions covered under one contract and a vendor backed by a large public telecom parent.
9. Appen — 30 Years Of Training Data With 1M+ Contributors In 235+ Languages
Appen is the longest-running company on this list. It was founded in Sydney in 1996, is listed on the ASX, and keeps a U.S. headquarters in Kirkland, Washington, under CEO Ryan Kolln. Appen reports more than 1 million vetted contributors in 170+ countries and support for 235+ languages. For the six months to June 30, 2026, revenue rose 17.5% to AUD 119.9 million, driven by an 80.4% jump in its China business, while Appen Global revenue fell 26.9% in a project-based, second-half-weighted segment, according to Staffing Industry Analysts.
Appen covers both sides of the Scale comparison. For LLM work it offers supervised fine-tuning demonstrations, RLHF and preference rankings from subject matter experts in medicine, law, science and finance, adversarial red teaming, chain-of-thought traces, hallucination detection and bias assessment. For classic annotation it handles LiDAR and sensor fusion labeling for physical AI, VLM annotation, and speech corpora and voice agent evaluation, run through its AI Data Platform (ADAP). Appen states that it is SOC 2 and ISO 27001 certified.
Appen’s main advantage is language coverage. Teams building speech recognition, multilingual chat or search relevance systems that need native speakers in dozens of locales will find few vendors with comparable reach. Customers shown on its site include Amazon, Microsoft, Nvidia, Adobe and Boeing. Buyers should still ask which delivery region and team will staff their project, given how differently its China and global segments are performing.
10. Sama — ISO 27001 Certified B Corp With A 99% First-Batch Acceptance Rate
Sama was founded in 2008 by Leila Janah as the nonprofit Samasource and became a for-profit company in 2019. It is headquartered in San Francisco and runs much of its delivery from East Africa, including a center in Nairobi. Sama is a Certified B Corporation and bases its quality and information security management system on ISO 9001 and ISO 27001.
Sama is a computer vision specialist. It annotates images, video and 3D point clouds from LiDAR and radar, and also offers GenAI annotation, model evaluation and data validation. The company reports a 99% first-batch acceptance rate and an average customer tenure of eight years. Clients and partners shown on its site include Continental, Getty Images, Qualcomm, Sony, Siemens, Walmart and NASA, and it publishes case studies with Walmart and Volumental. For more vendors in this category, see our computer vision annotation companies list.
Buyers should know about recent events. In early 2026, Meta ended its contract with Sama after Swedish newspapers reported that Nairobi workers had reviewed intimate footage from Ray-Ban Meta glasses. Meta said Sama did not meet its standards, a description Sama rejected. Sama then issued redundancy notices to 1,108 employees in April 2026, and Kenya’s data protection regulator opened an investigation, as The Next Web reported. Teams handling sensitive imagery should review Sama’s current data handling controls and project staffing directly.
How To Choose A Scale AI Alternative
Which Part Of Scale’s Work Are You Replacing?
Scale sold two different things: post-training data for frontier models and high-volume annotation for perception systems. Few vendors are equally strong at both. If your spend went to RLHF, demonstrations and evals, start with Surge AI, Mercor, Turing, Snorkel AI and Invisible. If it went to LiDAR, sensor fusion, video and mapping, start with iMerit, TELUS Digital, Appen and Sama. Labelbox sits in between because you can run either kind of work on its platform. Our list of RLHF and LLM training data companies goes deeper on the first group.
Who Owns The Vendor, And Does That Matter To You?
The Meta stake is why many teams left Scale, so check ownership and investor rosters at every alternative. Surge was bootstrapped through 2025, iMerit now belongs to EXL, TELUS Digital is wholly owned by TELUS, and Appen is publicly listed. Venture-backed vendors publish investor lists you can review. If a competitor holds a meaningful stake or board seat, decide in advance whether contract terms, data isolation and audit rights are enough protection for you.
Can The Vendor Prove Its Security Posture?
Training protocols, prompts and evaluation rubrics are trade secrets for many labs. Ask for current SOC 2 Type II reports, ISO 27001 certificates and, for automotive work, TISAX labels. Ask how contractor devices are managed, whether work happens in secure facilities or remotely, and how the vendor handled its most recent incident. The 2026 Mercor breach and the Sama footage controversy both show that supply chain and workforce controls matter as much as the certificate on the website.
How Are Experts Vetted And Paid?
Quality in post-training data depends on who writes and ranks the outputs. Ask how the vendor tests domain expertise, what share of contributors hold advanced degrees in your field, how rework and agreement rates are tracked, and how contributors are paid. Pay matters for retention and quality. Mercor publishes an average expert rate, while other vendors will share figures under NDA. A vendor that can show calibration data from a paid pilot gives you much more to go on than a talent pool headline number.
What Does A Pilot Look Like?
Run a short, paid pilot on a representative slice of your data before signing an annual commitment. Define acceptance criteria up front, such as inter-annotator agreement, first-pass acceptance rate or evaluation score movement, and ask each finalist to hit them on the same task. Compare turnaround, how often questions come back, and how easily the vendor’s output plugs into your pipeline. If you are budgeting for the wider program, our guide to AI development cost covers where data spend fits.
Conclusion
Meta’s 49% stake turned Scale AI from a default choice into a strategic decision, and the market of Scale AI alternatives has matured to match. Surge AI, Mercor, Turing, Snorkel AI and Invisible Technologies now compete directly for frontier post-training and evaluation budgets, while iMerit, TELUS Digital, Appen and Sama offer the managed workforces, certifications and sensor data experience that perception teams relied on Scale for. Labelbox gives teams a third route: keep the platform in-house and bring experts in on demand.
The right choice depends on which half of Scale’s business you are replacing, how much vendor ownership and security history matter to you, and how each finalist performs on a paid pilot with your own data. Many teams will end up with two vendors, one for GenAI data and one for classic annotation, which also reduces the single-supplier risk that pushed them away from Scale in the first place.
