Surge AI

Surge AI (legal entity Surge Labs Inc., often stylized SurgeHQ) is an American data annotation and human evaluation company headquartered in San Francisco, California that supplies frontier artificial intelligence laboratories with the human-generated training data, preference comparisons, red-teaming examples, and reinforcement learning environments used to fine-tune large language models. Despite taking no outside investment for its first five years, Surge reportedly generated about 1.2 billion dollars of revenue in 2024, surpassing the roughly 870 million dollars reported by venture-backed rival Scale AI and ranking, by revenue, as the largest pure-play AI data company in the world that year. ^[4][7][27] Founded in 2020 by former Google, Twitter, Dropbox, and Facebook research scientist Edwin Chen, the company built its business around a network of vetted domain-expert contractors rather than commodity crowd labor, charging substantially higher prices than legacy annotation vendors in exchange for higher-quality outputs on complex tasks such as Reinforcement Learning from Human Feedback (RLHF), agentic task evaluation, and safety red-teaming. ^[1][2][3] The company has been profitable since close to its launch and, through mid-2026, remained entirely bootstrapped, a posture that Reuters, Bloomberg, and Forbes all highlighted as unusual for a business at its scale. ^[4][5][7]

In July 2025, Reuters and Bloomberg reported that Surge had hired advisors for its first external capital raise of up to one billion dollars, at a valuation that began near fifteen billion dollars in early talks and reached at least twenty-five billion dollars within weeks, with Forbes later citing figures as high as thirty billion dollars. ^[4][5][8] As of mid-2026, however, no round had been publicly announced as closed, and Surge continued to report that it had taken no outside investment. Forbes nonetheless estimated founder Edwin Chen's net worth at about eighteen billion dollars on his roughly 75 percent stake, ranking him No. 55 on the 2025 Forbes 400 as its youngest member and No. 1 on the magazine's December 2025 list of the richest self-made billionaires under forty. ^{[8][27][28][29]}

How did Surge AI start, and who is Edwin Chen?

Edwin Chen's background

Edwin Chen (born 1987 or 1988) is a graduate of the Massachusetts Institute of Technology, where he studied mathematics, computer science, and linguistics. ^[3][13] The son of Taiwanese immigrants who ran a family restaurant, Chen attended the Choate Rosemary Hall boarding school, the final two years on a full scholarship, before entering MIT, and he now lives in Manhattan. ^[27][28] Following MIT he spent roughly a decade as a research scientist and applied machine learning engineer at major technology platforms, including Google, Twitter, Dropbox, and Facebook (now Meta), as well as Peter Thiel's hedge fund Clarium Capital, working on areas such as search, recommendation systems, content moderation, and content understanding. ^[3][13][28] During these years he also maintained a widely read technical blog at echen.me on topics including Bayesian inference, latent variable models, recurrent neural networks, and recommender systems. ^[13][42]

According to interviews Chen has given to Inc. and Lenny's Newsletter, the immediate motivation for founding Surge was the gap he observed between the increasingly sophisticated machine learning systems being deployed by major labs and the comparatively crude human annotation pipelines feeding them, in which low-paid crowd workers produced noisy labels for complex linguistic and reasoning tasks. ^[2][3] Chen left his last job, at Twitter, in 2020 and founded Surge that year as the COVID-19 pandemic began, funding it with his own savings ("a couple million" dollars from his decade in Big Tech, according to Forbes), and deliberately chose not to raise venture capital despite the prevailing Silicon Valley orthodoxy. ^[27][42]

Surge's early years and product strategy

Surge's earliest work was concentrated in natural language data: search relevance evaluation, content moderation, recommendation quality, and similar projects for product teams at companies that had previously been Chen's employers and adjacent labs. ^[2][3] The company's foundational thesis was that the quality of model outputs is bounded by the quality of the human data used to train and evaluate them, and that the prevailing low-cost crowdsourcing model (typified by Amazon Mechanical Turk and Eastern European platforms such as Toloka) was inadequate for the level of nuance required by post-2020 language modeling. ^[3][14]

In response, Surge built what Sacra describes as a "managed marketplace" of vetted domain experts ("Surgers") who could handle complex tasks in mathematics, coding, law, medicine, science, and creative writing. ^[10][12] The company implemented proprietary quality control systems that combined inter-annotator agreement metrics, gold-standard test items, per-worker trust scores, and machine-aided checks, with low-quality outputs flagged and reassigned. ^[11][12]

By 2022 the company, profitable almost from its launch, had quietly become a primary data supplier for several frontier AI labs, including Anthropic, whose co-founder Jared Kaplan endorsed Surge's RLHF platform in a 2023 testimonial published on Surge's site. ^[12] As the post-ChatGPT boom drove demand for RLHF data, Surge expanded heavily into preference comparison work (in which annotators rank multiple model responses), red-teaming, and safety evaluation. ^[12][14]

How Surge grew and gained public visibility

Surge AI operated with deliberately low public visibility through 2024. The company had no traditional sales team, no chief revenue officer, and Chen himself was largely absent from media coverage and conference circuits, with one widely cited summary noting that even after his name surfaced on Forbes lists his LinkedIn profile read only "Building Surge AI." ^[15]

The first wave of mainstream coverage came in mid-2025. In July 2025, Reuters reported, citing people familiar with the matter, that Surge had hired financial advisors to raise as much as one billion dollars in what would be the firm's first capital round, with an initial valuation target above fifteen billion dollars, and that the company had cleared more than one billion dollars of revenue in 2024, exceeding the approximately 870 million dollars reported that year by venture-backed rival Scale AI. ^[4][16] The Reuters reporting was carried by Yahoo Finance, U.S. News and SiliconANGLE, among other outlets. ^[4][16][17] Within weeks Bloomberg reported that the talks had progressed to a target valuation of at least twenty-five billion dollars. ^[5]

Forbes followed in September 2025 with a profile by Phoebe Liu reporting that Surge's 2024 revenue was approximately 1.2 billion dollars, that the company had been profitable since near inception, and that Chen, retaining roughly seventy-five percent of equity, had a paper net worth of about eighteen billion dollars on the basis of the funding-round valuation, entering that year's Forbes 400 at No. 55 and, at age thirty-seven, ranking as its youngest member. ^{[7][8][27][28]} TIME named Chen to its TIME100 AI 2025 list the same year, noting that Surge sells its datasets to companies such as Google, Anthropic and OpenAI. ^[18]

Industry tracker Sacra estimated that Surge's 2024 revenue of about 1.2 billion dollars came mainly from roughly twelve frontier AI lab customers, served by approximately 130 full-time employees and a global workforce of about fifty thousand expert contractors. ^[10] Forbes, counting part-time staff and consultants, put the headcount closer to 250, still a very small team for the revenue involved. ^[27] Those figures imply revenue of roughly nine million dollars per full-time employee, an unusual ratio for a labor-intensive services business. ^[10]

How does Surge AI's business model work?

Why does Surge use vetted domain-expert labelers?

Surge's central differentiator is its workforce model. Whereas commodity annotation platforms typically distribute simple tasks to large pools of low-wage workers (often based outside the United States) at rates of a few dollars per hour, Surge recruits and screens contractors with verifiable subject-matter expertise, including practicing lawyers, physicians, mathematicians, software engineers, and professional writers, and reportedly pays rates well above standard crowdsourcing levels, which Sacra put at 30 to 40 cents per working minute (about 18 to 24 dollars per hour). ^[10]

Surge vets its contractors with domain-specific tests, background checks, and ongoing performance evaluations. ^[10] Active contractors are continuously monitored against gold-standard items, inter-annotator agreement, and project-specific quality dashboards, and low-quality submissions are reassigned to other workers before being returned to the customer. ^[11][12] This pipeline allows Surge to provide annotations for tasks such as evaluating mathematical proofs, debugging code, ranking legal arguments, or assessing the safety of generated text, where commodity labelers cannot reliably perform. ^[12][14]

What kinds of data does Surge produce?

Surge's product surface is organized around several categories of human data used by large language model developers in post-training:

• Preference comparisons for RLHF, in which annotators rank or score multiple model outputs against each other on dimensions such as helpfulness, harmlessness, factuality, or stylistic quality. These pairs form the basis for training reward models that are subsequently used as the reinforcement signal during policy optimization. ^[12][14] Surge has been an explicit partner on RLHF data for Anthropic's Claude family, with Anthropic co-founder Jared Kaplan publicly describing Surge as "an excellent partner" in technical AI alignment research. ^[12]
• Supervised fine-tuning demonstrations, in which expert labelers write high-quality target responses to prompts in their area of expertise, used in supervised fine-tuning passes before reinforcement learning. ^[12]
• Evaluation and benchmarking, in which experts grade or compare model outputs on tasks such as mathematics, coding, factual question answering, and complex reasoning. OpenAI's GSM8K paper credits the team at Surge AI with performing the data collection for that grade-school math benchmark. ^[41] Surge also publishes benchmarks of its own. On April 14, 2026 it announced GDP.pdf, a benchmark of 100 real-world prompts and PDFs (contracts, invoices, medical records, blueprints and similar documents) from ten professional domains, and said every frontier model scored under 15%. ^[36][40] The paper by Suhaas Garre, Emily Ritchie, Sushant Mehta and Edwin Chen was posted to arXiv (2607.11192) on July 13, 2026. ^[37] OpenAI listed GDP.pdf in its GPT-5.6 release in July 2026 and in its September 29, 2026 GPT-6.1 Sol launch post, and Anthropic has reported results from its own internal GDP.pdf harness in system cards, including the one for Claude Sonnet 5. ^[38][39][40] On July 16, 2026 Surge released Chartography, a benchmark of professional chart understanding built on specialized graphics such as Kaplan-Meier curves, Sankey diagrams and Bode plots. The accompanying paper by Suhaas Garre, Chris Mutty, Sushant Mehta and Edwin Chen, submitted to arXiv on August 11, 2026, describes 100 tasks with questions written by professionals who read these charts for a living, and reports that the best of 30 frontier-model configurations reached 45.0% mean pass@1. ^[34][35]
• Red-teaming and adversarial data, in which annotators with creative writing, security, or AI/ML backgrounds attempt to elicit unsafe, biased, or otherwise undesirable model behavior. These adversarial examples, paired with model refusals or corrected behavior, feed into AI alignment training, including frameworks akin to Constitutional AI and refusal-tuned InstructGPT-style pipelines. ^[12][20]
• Reinforcement learning environments for agentic tasks, in which Surge labelers design or evaluate multi-step task trajectories used to train agents on tool use, browsing, code execution, and similar workflows. Edwin Chen has publicly identified reinforcement learning environments as the next frontier in AI training data. ^[3][10]

Pricing is reportedly project-based or usage-based, with Surge frequently charging multiples of competitor pricing. Forbes reported that Surge charges between roughly 50 percent and ten times more than competitors, a premium the company justifies by the customer-perceived gap in output quality on frontier-relevant tasks. ^[14][19][27]

Surge's labeling platform and toolchain

Surge operates a proprietary labeling platform that exposes APIs and web interfaces for customers to define tasks, define rubrics, manage gold standards, view live quality dashboards, and integrate outputs into model training pipelines. ^[12] The platform supports rapid experimentation, enabling a model team to launch a new annotation campaign within hours rather than days, and supports highly structured rubrics for nuanced judgments.

How does Surge grow without a sales team?

A repeated theme in coverage of Surge is its growth without conventional outbound sales. Multiple commentators have described the company as relying on a "researcher flywheel" in which individual machine learning researchers who have used Surge data at one lab introduce the vendor to subsequent labs as they change employers; given the relatively small population of senior frontier-lab researchers, the network effect is rapid. ^[3][10][19] Chen has publicly framed his rejection of venture capital in similar terms, telling Inc. that VC funding induces "politics," "bureaucracy," and a cycle of "you raised ten million dollars, what are you going to do with that money?" pressure that he believed would have degraded the product and the culture of the company. ^[2][10] "I've always hated the Silicon Valley status game," Chen told Forbes, framing his avoidance of fundraising, press, and the conference circuit as of a piece with a focus on the work itself. ^[27]

Who are Surge AI's customers?

Public reporting and Surge's own published case studies have identified the organizations below as customers, with two caveats: several of these contracts are described in secondary reporting and have not been confirmed in writing by Surge, and the roster shifted during 2025 as some labs moved work between vendors.

The company has occasionally published joint case studies with customers, for example a Surge blog post in 2023 describing Anthropic's use of the Surge RLHF platform for training the Claude assistant, which included testimonial language from Anthropic's Jared Kaplan. ^[12] Surge has also publicly described work with Redwood Research on adversarial data labeling for AI safety research. ^[20] Forbes reported that Surge's data has helped train systems including Google's Gemini and Anthropic's Claude. ^[27]

How does Surge AI compare with Scale AI, Mercor, and other data companies?

The post-2022 surge in demand for human-generated training data, particularly for RLHF and evaluation of frontier language models, gave rise to a competitive landscape with several distinct strategic clusters. The table below summarizes how Surge compares with frequently cited peers; figures are drawn from secondary reporting and may not be directly comparable methodologically.

Several themes recur in the comparative coverage of these vendors. First, the rise of Surge and Mercor is widely framed as a quality-driven backlash against the commodity model exemplified by older players such as Sama, Appen, and Cloudfactory, which had built their workforces around low-cost geographies and now struggle to provide the expertise required for frontier-lab RLHF. ^[14] Second, customer churn from Scale AI after Meta's June 2025 investment (with Google, OpenAI, and xAI reportedly winding down Scale work for data-security reasons) is widely reported to have accelerated growth at Surge and Mercor. ^[14][21] Third, internal Meta researcher feedback reportedly favors Surge and Mercor over Scale for RLHF data quality, a notable reversal given Meta's ownership stake in Scale. ^[14]

By mid-2026 the competitive picture had sharpened. Meta's June 2025 investment in Scale, a 49 percent non-voting stake worth 14.3 billion dollars that valued Scale at nearly 29 billion dollars and moved founder Alexandr Wang to Meta's superintelligence effort (with Jason Droege promoted to Scale's chief executive), prompted several labs to reduce their reliance on Scale and to shift work toward independent vendors including Surge and Mercor. ^[31] Mercor grew especially fast: in July 2026 founder Brendan Foody said its annualized revenue run rate had crossed two billion dollars, double the level of four months earlier, and Bloomberg reported that it was in early talks to raise at a valuation of about 20 billion dollars, roughly double the 10 billion dollars of its October 2025 Series C. ^[32] The same reshuffling that benefited Surge also cost it a marquee account: Forbes reported that OpenAI, once a Surge customer, had stopped using the company and moved annotation work to rivals such as Mercor and Invisible Technologies. ^[27]

Why does Surge AI matter for frontier AI training?

Surge's economic significance is a direct consequence of the rise of RLHF and related techniques as the dominant post-training paradigm for frontier language models. After OpenAI's InstructGPT paper (2022) demonstrated that aligning a base large language model with human preference data produced dramatic improvements in helpfulness and harmlessness, the dominant frontier labs adopted RLHF and related preference-based methods (including Direct Preference Optimization (DPO) and Constitutional AI) as the core stage between pre-training and deployment. ^[22] The bottleneck of these pipelines shifted from raw compute and pre-training data to the supply of high-quality, expert-generated comparison and demonstration data, which created the market in which Surge competes. ^[14][22]

In coverage of frontier model releases since 2023, Surge has been repeatedly characterized as a key behind-the-scenes contributor to model quality. In a Lenny's Newsletter interview published in late 2025, Chen and the show framed Surge's data work as having materially contributed to capabilities including Claude Code's coding and writing performance, and described the company as a "secret weapon" supplying multiple major labs. ^[3] Chen has made expansive claims about the company's role, telling Forbes bluntly that "without us, AGI just won't happen." ^[27] Surge's own corporate framing rejects the term "data labeling" outright: Chen has stated in interviews that he "always hated the word data labeling" because it understates the substantive judgment required, and he frequently compares the work to raising a child, in which the goal is not merely to feed information but to instill values, taste, and creativity. ^[23]

From an industry-structure perspective, Surge's bootstrapped growth to revenue parity with Scale AI is often cited as a counterexample to the prevailing assumption that AI infrastructure businesses require billions of dollars of venture capital. Multiple commentators have framed Surge as among the most capital-efficient companies in Silicon Valley history, with Lenny's Newsletter describing it as the fastest company in history to reach one billion dollars in revenue. ^[3] The company's labor model also positions it as a counterpoint to concerns about offshore "AI ghost work," since its labelers are described as vetted subject-matter experts paid above crowdsourcing rates. ^[10]

What controversies and limitations does Surge AI face?

Worker misclassification lawsuit (2025)

In May 2025, Bloomberg Law and other outlets reported that Surge AI had been named in a putative class action alleging that the company misclassified its data annotators as independent contractors and, as a result, denied them protections owed to employees under California labor law, including minimum wage, overtime, and meal and rest break protections. The complaint, filed on May 20, 2025 in California Superior Court in San Francisco by plaintiff Dominique DonJuan Cavalier II and brought by the public-interest Clarkson Law Firm, alleges that annotators were required to perform unpaid training, were subjected to deadlines and supervision incompatible with contractor status, and worked on training projects for major Surge customers including Meta and OpenAI. ^[24][25][33] The litigation parallels similar actions against Scale AI and Mercor and is part of a broader pattern of legal scrutiny applied to AI training labor that continued into 2026. ^[25][32]

Questions about affiliated worker platforms

Reporting on the AI data labor sector has raised questions about several worker-facing platforms that appear to be operated by Surge. The Verge and New York Magazine reported in 2023 that Surge appeared to own multiple separate labeling platforms marketed to workers under different names, including DataAnnotation.tech, Gethybrid.io, and Taskup.ai, without clearly disclosing the common ownership. ^[30] The May 2025 class-action complaint names the defendant as "Surge Labs, Inc. dba DataAnnotation." ^[33] Surge has not published a comprehensive account of its subsidiary platform structure, and several commentators have flagged the ambiguity as a transparency concern. Workers on these platforms have separately reported unexplained account terminations and limited recourse, which Surge has not publicly addressed in detail. ^[30]

Leaked internal safety guidelines (2025)

In July 2025, copies of internal Surge documents describing training-data guidelines were reported to have leaked publicly. One document covered the handling of sensitive content and AI-safety-related instructions, and another reportedly detailed which websites contractors could and could not use when generating data for Anthropic model training. ^[6][30] Coverage of the leak noted that decisions made by data vendors and their guideline writers, often in the absence of customer-visible disclosure, can materially shape downstream model behavior, raising governance questions about how such decisions should be reviewed. ^[6]

Customer concentration

Industry trackers have noted that Surge's revenue base is highly concentrated, with approximately twelve large frontier AI lab customers reportedly accounting for the majority of revenue. ^[10] This concentration creates exposure to demand swings driven by individual customer roadmap decisions, customer-side build-versus-buy choices, or competitive substitution by rival vendors such as Mercor or in-house data teams. The risk is not hypothetical: Forbes reported that OpenAI, once a customer, had moved its annotation work elsewhere by late 2025. ^[10][14][27]

Can the premium pricing last?

Some industry analysts have questioned whether Surge's premium-priced, expert-driven model can sustain its pricing as competitors (including Mercor and revived offerings from larger annotation vendors) catch up on quality, and as customer labs increasingly invest in internal data infrastructure, synthetic data pipelines, and automated AI-on-AI evaluation. Equidam and other analyst commentary in 2025 noted that Surge's valuation, while supported by strong fundamentals, is sensitive to whether high-quality human data continues to command premium pricing over the next several years. ^[19]

Has Surge AI raised funding, and will it go public?

Through the first half of 2025 Surge AI had taken no external capital beyond Chen's initial self-financing, an unusual posture for a company at its scale. ^[2][3][10] In July 2025 Reuters reported that Surge had engaged advisors to raise as much as one billion dollars in a mixed primary and secondary round, targeting an initial valuation of at least fifteen billion dollars. ^[4][16] By late July 2025 Bloomberg reported that talks had advanced to a valuation of at least twenty-five billion dollars, with potential investors including Andreessen Horowitz, Warburg Pincus, and TPG Inc. ^[5][9] Forbes in September 2025 reported that talks were proceeding at a valuation as high as thirty billion dollars, and used a figure of roughly twenty-four billion dollars in estimating Chen's net worth; the company's valuation trajectory across 2025 was widely tracked in the trade press. ^{[6][7][8][27]}

As of November 2026, no funding round had been publicly announced as completed. Forbes' profile of Chen, last updated in September 2026, still noted that "Surge claims it has not yet raised any outside funding," and the magazine's December 2025 "40 Under 40" list likewise described him as having "grown Surge without raising any outside funding." ^[28][29] Public commentary has continued to speculate about a potential initial public offering. Surge has not announced IPO plans, and Chen has repeatedly downplayed the idea, telling Inc. only "Who knows what will happen in the future?" and telling Forbes, "Why would anyone want to go public? A big problem with public companies is they always have to worry about the short term." ^[2][27] Equidam's 2025 valuation analysis estimated that an investor entering at 25 billion dollars would need an exit of roughly 45 to 50 billion dollars, using Coinbase's profitable 2021 IPO as a point of comparison. ^[19]

Related work

Surge's business intersects with several related areas of artificial intelligence research and infrastructure:

• Reinforcement Learning from Human Feedback (RLHF) is the dominant post-training paradigm that Surge's preference data feeds.
• Reinforcement learning from human feedback provides a longer treatment of the underlying technique.
• Constitutional AI is the Anthropic-developed alignment approach that augments human preference data with model-self-critique and is among the methods that Surge's human feedback supports.
• Direct Preference Optimization (DPO) and other related algorithms substitute for the RL step in classical RLHF and consume the same preference comparison data that Surge produces.
• Supervised fine-tuning is the earlier pipeline stage in which expert-written demonstration data, of the kind Surge supplies, is used to instruct the model.
• Post-training is the umbrella term for the pipeline stages in which Surge's products are predominantly used.
• AI Alignment is the broader research field whose practical implementation depends heavily on the quality of human feedback infrastructure such as Surge's.
• Red teaming (artificial intelligence) describes the adversarial evaluation practice that Surge supports.
• Data labeling is the legacy framing of the work Surge does, though Chen has consistently rejected the term.

Continue reading

• Reinforcement Learning from Human Feedback (RLHF)Reinforcement learning from human feedback (RLHF) is a family of methods in which judgments supplied by people are converted into a reward signal and a policy is optimized with reinforcement learning against…Mentioned 22 times above
• OpenAIOpenAI is an American artificial intelligence research and deployment organization.Mentioned 13 times above
• MercorMercor is an AI-powered hiring and talent marketplace that connects vetted human domain experts, including engineers, lawyers, doctors, bankers, and journalists, with frontier AI laboratories that need…Mentioned 12 times above
• AnthropicAnthropic is an American artificial intelligence (AI) safety and research company founded in 2021 by Dario Amodei, Daniela Amodei, and other former OpenAI researchers, best known for the Claude family of large…Mentioned 12 times above
• Meta AIMeta AI is the name Meta Platforms uses for two related but distinct things: the company's artificial intelligence research and engineering organizationMentioned 11 times above
• Scale AIScale AI is a San Francisco-based artificial intelligence company founded in 2016 by Alexandr Wang and Lucy Guo.Mentioned 8 times above

More from See also

• Data labeling
• Constitutional AI
• InstructGPT
• Direct Preference Optimization (DPO)
• Supervised fine-tuning
• Post-training
• AI Alignment
• Red teaming (artificial intelligence)
• Google
• Microsoft
• Mistral AI
• Cohere
• Snorkel
• Redwood Research
• Claude (language model)
• Claude Code
• GSM8K
• Large Language Model
• Frontier Models
• Andreessen Horowitz
• Massachusetts Institute of Technology

Topics:AI CompaniesData & Datasets

References

1. ^1 ^2 ^3 ^4Surge AI, "Surge AI: Powering Frontier AI With High-Quality Human Data" (company site), surgehq.ai, 2025. surgehq.ai Accessed 2026-05-20.
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Brian Contreras, "Surge AI Bootstrapped Its Way to $1 Billion in Revenue", Inc., 2025-07. inc.com/...91205888 Accessed 2026-05-20.
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13Lenny Rachitsky, "The 100-person AI lab that became Anthropic and Google's secret weapon | Edwin Chen (Surge AI)", Lenny's Newsletter, 2025. lennysnewsletter.com/...surge-ai-edwin-chen Accessed 2026-05-20.
4. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Reuters (Milana Vinn and Krystal Hu), "Exclusive: Scale AI's bigger rival Surge AI seeks up to $1 billion capital raise, sources say", reported via Yahoo Finance, 2025-07-01. finance.yahoo.com/...le-ais-bigger-rival-152327620 Accessed 2026-05-20.
5. ^1 ^2 ^3 ^4 ^5 ^6Bloomberg News, "Surge AI in Talks for Funding at $25 Billion Value", Bloomberg, 2025-07-30. bloomberg.com/...s-for-funding-at-25-billion-value Accessed 2026-05-20.
6. ^1 ^2 ^3AI CERTs News, "Surge AI's Data Labeling Valuation Saga", AI CERTs News, 2025. aicerts.ai/...surge-ais-data-labeling-valuation-saga Accessed 2026-05-20.
7. ^1 ^2 ^3 ^4 ^5 ^6Phoebe Liu / Forbes (via Techmeme aggregation), "A profile of Edwin Chen, the CEO of Surge AI, a Scale AI rival that had $1.2B revenue in 2024, is profitable, and is reportedly raising $1B at a $30B valuation", Techmeme, 2025-09-21. techmeme.com/...p5 Accessed 2026-05-20.
8. ^1 ^2 ^3 ^4 ^5VnExpress International, "Meet Edwin Chen, MIT graduate and youngest billionaire on Forbes' 400 richest Americans list", VnExpress International, 2025. e.vnexpress.net/...-richest-americans-list-4938703 Accessed 2026-05-20.
9. ^Reuters (via U.S. News), "Exclusive-Scale AI's Bigger Rival Surge AI Seeks up to $1 Billion Capital Raise, Sources Say", U.S. News & World Report, 2025-07-01. money.usnews.com/...lion-capital-raise-sources-say Accessed 2026-05-20.
10. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18Sacra, "Surge AI revenue, funding & news", Sacra, 2025. sacra.com/...surge-ai Accessed 2026-05-20.
11. ^1 ^2 ^3 ^4 ^5Skywork.ai, "Surge AI: The Ultimate Guide for AI Practitioners", Skywork.ai, 2025. skywork.ai/...1976160605372608512 Accessed 2026-05-20.
12. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15Surge AI Blog, "Anthropic uses Surge AI's RLHF platform to train LLM Assistant with human feedback", surgehq.ai, 2023 (republished 2025). surgehq.ai/...m-train-llm-assistant-human-feedback Accessed 2026-05-20.
13. ^1 ^2 ^3Edwin Chen, "Edwin Chen | Surge AI" (personal/professional site with biography and archived blog), edwinchen.ai, 2025. edwinchen.ai Accessed 2026-05-20.
14. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15Ishi Singhal, "The Evolution of Data Labelling: Sama, Scale AI, Surge AI and Mercor", Medium, 2025. medium.com/...-ai-surge-ai-and-mercor-8a8d69514336 Accessed 2026-05-20.
15. ^36Kr (Europe edition), "37-Year-Old Genius Chinese-American Becomes the Youngest Billionaire", 36Kr, 2025. eu.36kr.com/...3502837065586825 Accessed 2026-05-20.
16. ^1 ^2 ^3SiliconANGLE (Mike Wheatley), "Data labeling startup Surge AI reportedly seeking $1B in first capital raise", SiliconANGLE, 2025-07-01. siliconangle.com/...irst-capital-raise-reports-say Accessed 2026-05-20.
17. ^Yahoo Finance, "Surge AI Quietly Hit $1B Without Outside Money, Now Even VCs Want In", Yahoo Finance, 2025-07-19. finance.yahoo.com/...e-ai-quietly-hit-1b-150057861 Accessed 2026-05-20.
18. ^TIME, "Edwin Chen: The 100 Most Influential People in AI 2025", TIME (TIME100 AI 2025), 2025. time.com/...edwin-chen Accessed 2026-05-20.
19. ^1 ^2 ^3 ^4 ^5Equidam, "AI Valuation Explained: Surge AI vs Scale AI", Equidam, 2025. equidam.com/ai-valuation-scale-ai-vs-surge-ai Accessed 2026-05-20.
20. ^1 ^2Surge AI Blog, "AI Red Teams and Adversarial Data Labeling, with Redwood Research", surgehq.ai, 2023. surgehq.ai/...-data-labeling-with-redwood-research Accessed 2026-05-20.
21. ^1 ^2 ^3 ^4Jennifer Conrad, "How Surge AI Is Already Outpacing Rival Scale AI", Inc., 2025. inc.com/...91204563 Accessed 2026-05-20.
22. ^1 ^2Long Ouyang et al., "Training language models to follow instructions with human feedback (InstructGPT)", arXiv:2203.02155, 2022-03-04. arxiv.org/...2203.02155 Accessed 2026-05-20.
23. ^AOL (republishing Business Insider / Inc.), "Surge AI CEO explains why he hates the term 'data labeling'", AOL, 2025. aol.com/...surge-ai-ceo-explains-why-100002672 Accessed 2026-05-20.
24. ^Bloomberg Law, "AI Training Firm Surge AI Hit With Worker Misclassification Suit", Bloomberg Law, 2025-05. news.bloomberglaw.com/...er-misclassification-suit Accessed 2026-05-20.
25. ^1 ^2Worksuite, "What Companies Can Learn from the Surge AI and Scale AI Lawsuits", Worksuite, 2025. worksuite.com/...ai-misclassification-lawsuits Accessed 2026-05-20.
26. Sam Blum, "Bootstrapped to $1 Billion: Surge AI CEO Edwin Chen on How He Did It", Inc., 2025. inc.com/...91207937 Accessed 2026-05-20.
27. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24 ^25 ^26Phoebe Liu, "How The Low-Key Billionaire Behind Surge Is Beating Out Rivals Like Scale AI" (also published as "The AI Billionaire You've Never Heard Of"), Forbes, 2025-09-17. forbes.com/...-ai-billionaire-youve-never-heard-of Accessed 2026-07-12.
28. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Forbes, "Edwin Chen" (real-time billionaire profile: net worth, No. 55 on the 2025 Forbes 400, No. 158 on the 2026 Billionaires list, roughly 75 percent ownership), Forbes, updated 2026-07-11. forbes.com/...edwin-chen Accessed 2026-07-12.
29. ^1 ^2Matt Durot, "40 Under 40: The Richest Self-Made Billionaires Under 40", Forbes, 2025-12-29. forbes.com/...hest-self-made-billionaires-under-40 Accessed 2026-07-12.
30. ^1 ^2 ^3 ^4"Surge AI", Wikipedia, 2026 (summarizing 2023 reporting by The Verge and New York Magazine on affiliated worker platforms, and the 2025 document leaks). en.wikipedia.org/...Surge_AI Accessed 2026-07-12.
31. ^1 ^2CNBC, "Scale AI's Alexandr Wang confirms departure for Meta as part of $14.3 billion deal", CNBC, 2025-06-12. cnbc.com/...-exit-for-meta-part-of-14-billion-deal Accessed 2026-07-12.
32. ^1 ^2 ^3 ^4TechCrunch, "Mercor is in talks for a $20B valuation", TechCrunch, 2026-07-09. techcrunch.com/...-is-in-talks-for-a-20b-valuation Accessed 2026-07-12.
33. ^1 ^2Clarkson Law Firm, "Class Action Complaint, Cavalier v. Surge Labs, Inc." (California Superior Court, San Francisco), filed 2025-05-20. clarksonlawfirm.com/...2025.05.20-Surge-Labs.pdf Accessed 2026-07-12.
34. ^Surge AI Blog, "Chartography: A Benchmark for Professional Chart Understanding", surgehq.ai, 2026-07-16. surgehq.ai/...chartography Accessed 2026-09-29.
35. ^Garre, Suhaas; Mutty, Chris; Mehta, Sushant; Chen, Edwin, "Chartography: A Benchmark for Professional Chart Understanding", arXiv:2608.10677, 2026-08-11. arxiv.org/...2608.10677 Accessed 2026-09-29.
36. ^Surge AI (@HelloSurgeAI), "Introducing GDP.pdf: an expert multimodal reasoning benchmark for the documents that run the world", X, 2026-04-14. x.com/...2044115915718177208 Accessed 2026-09-30.
37. ^Garre, Suhaas; Ritchie, Emily; Mehta, Sushant; Chen, Edwin, "GDP.pdf: Benchmarking Grounded Multimodal Reasoning over Professional PDF Documents", arXiv:2607.11192, 2026-07-13. arxiv.org/...2607.11192 Accessed 2026-09-30.
38. ^Surge AI Blog, "OpenAI cites GDP.pdf in its GPT-5.6 release", surgehq.ai, 2026-07-09. surgehq.ai/...openai-gpt-5-6-gdp-pdf-benchmark Accessed 2026-09-30.
39. ^OpenAI, "Introducing GPT-6.1 Sol", openai.com, 2026-09-29. openai.com/...introducing-gpt-6-1-sol Accessed 2026-09-30.
40. ^1 ^2Anthropic, "Claude Sonnet 5 System Card", section 8.10.1, 2026-06-30. anthropic.com/claude-sonnet-5-system-card Accessed 2026-09-30.
41. ^1 ^2Karl Cobbe et al., "Training Verifiers to Solve Math Word Problems" (GSM8K), arXiv:2110.14168, 2021-10-27. arxiv.org/...2110.14168 Accessed 2026-11-05.
42. ^1 ^2Edwin Chen, "Edwin Chen's Blog" (including "Surge: Data Labeling You Can Trust", 2020), blog.echen.me, archived 2020. web.archive.org/...blog.echen.me Accessed 2026-11-05.

Show all 42 references