Mercor評測：AI資料標注新創估值半年翻倍衝200億 | Mercor Review: AI Data Labeling Startup Doubles to $20B
By Kit 小克 | AI Tool Observer | 2026-08-20
🇹🇼 Mercor評測：AI資料標注新創估值半年翻倍衝200億
Mercor是這波AI資料標注（AI Data Labeling）熱潮裡最受矚目的新創之一，最新傳出正在洽談以200億美元估值募資5億美元，比九個月前C輪的100億美元估值直接翻倍，連Nvidia都在考慮跟投。這篇文章帶你搞懂Mercor在做什麼、為什麼它能漲這麼快，以及這波熱潮跟一般AI工具使用者有什麼關係。
Mercor是什麼？AI資料標注新創在做什麼？
Mercor是一個媒合平台，讓醫生、律師、工程師等領域專家幫OpenAI、Anthropic、Nvidia等公司產生高品質訓練資料，用來提升AI模型的推理與專業能力。
Mercor最早在2023年以「AI徵才媒合」起家，創辦人Brendan Foody、Adarsh Hiremath、Surya Midha都是輟學的Thiel Fellow。後來公司發現真正的商機不是幫企業找人，而是幫AI實驗室找「懂專業的人」來標注資料、寫評測題、做人類回饋強化學習（RLHF），於是全面轉型。這個轉型讓Mercor從一年營收約1億美元，衝到目前年化營收突破20億美元，成長速度相當驚人。
為什麼Nvidia想投資Mercor？
Nvidia自己也是Mercor的客戶，用它的人類標注資料訓練自家開源模型，投資等於是綁住重要的資料供應商，同時分享AI基礎設施上下游的成長紅利。
這種「晶片廠投資資料供應商」的操作並不新鮮，但反映出一個現實：算力（Nvidia GPU）跟資料（Mercor的人類標注）已經是AI競賽的兩大瓶頸，誰卡住上游誰就有議價權。
Mercor的估值為什麼半年翻倍到200億美元？
核心原因是各大AI實驗室對「高品質、專業領域」訓練資料的需求爆炸性成長，而Mercor靠平台化媒合把交付速度做得比傳統標注公司快很多。
領投方General Catalyst加碼、既有客戶OpenAI與Anthropic持續加單，讓Mercor的營收曲線非常陡峭，這也是矽谷願意用兩倍估值追加碼的主因。三位創辦人若這輪順利完成，帳面身價各自逼近43億美元。
Kit小克怎麼看：這波熱潮跟你我有關係嗎？
老實說，一般人不會直接用到Mercor這種平台，但它解釋了一件事：你每天在用的ChatGPT、Claude回答變準、幻覺變少，背後很大一部分功勞來自這些看不見的人類標注員。
- 好處：專業領域資料品質提升，代表AI工具在醫療、法律、程式碼等專業場景的可靠度會持續進步。
- 要注意的地方：這類「資料標注獨角獸」估值狂飆，跟AI晶片股一樣有泡沫疑慮，一旦模型訓練需求放緩，這類公司的營收會很敏感地反應出來。
- 倫理面：零工經濟式的標注勞動報酬與工作條件，一直是這類公司被外界質疑的地方，Mercor也不例外。
對AI工具使用者來說，與其追這類募資新聞的熱度，不如觀察你常用的模型在你的專業領域是不是真的變準了——那才是資料標注投資有沒有真正轉化成產品品質的證據。
常見問題 FAQ
Q: Mercor跟Scale AI有什麼不同？
A: 兩者都是AI資料標注平台，Scale AI已被Meta大舉投資並挖角高層，Mercor則走獨立募資路線，客戶涵蓋OpenAI、Anthropic、Nvidia等多家實驗室。
Q: Mercor這輪估值200億美元確定了嗎？
A: 截至目前仍在洽談階段，尚未正式收官，200億美元是媒體引述消息來源的預估數字。
Q: 一般開發者可以用Mercor的服務嗎？
A: Mercor主要服務對象是大型AI實驗室與企業客戶，並非面向一般開發者的自助工具。
好不好用，試了才知道。
🇺🇸 Mercor Review: AI Data Labeling Startup Doubles to $20B
Mercor, an AI data labeling startup, is in talks to raise $500 million at a $20 billion valuation — exactly double its $10 billion Series C from just nine months ago. Nvidia is reportedly weighing a stake. This review breaks down what Mercor actually does, why its valuation is climbing so fast, and what it means for everyday AI tool users.
What Is Mercor and What Does an AI Data Labeling Startup Do?
Mercor connects domain experts — doctors, lawyers, engineers — with AI labs like OpenAI, Anthropic, and Nvidia to produce high-quality training data used for reinforcement learning from human feedback (RLHF).
Founded in 2023 as an AI-powered hiring marketplace by Thiel Fellowship dropouts Brendan Foody, Adarsh Hiremath, and Surya Midha, Mercor pivoted once it realized the real demand wasn't matching companies with hires — it was matching AI labs with credentialed experts who could grade model outputs and write specialized evaluation data. That pivot took Mercor from roughly $100M in annual revenue to a reported $2 billion annualized run rate today.
Why Is Nvidia Considering an Investment in Mercor?
Nvidia is already a Mercor customer, using its human-labeled data to train its own open-source models — investing would secure a key upstream data supplier while capturing upside from AI infrastructure growth.
Chipmakers investing in data suppliers isn't new, but it underscores a real bottleneck: compute (Nvidia GPUs) and data (Mercor's expert labeling) are now the two chokepoints of the AI race, and controlling either gives real leverage.
Why Did Mercor's Valuation Double to $20B in Six Months?
Demand for high-quality, domain-specific training data from frontier labs is exploding, and Mercor's platform model delivers it faster than legacy data-labeling firms. Existing investor General Catalyst is reportedly leading the new round, with OpenAI and Anthropic continuing to expand their orders — a revenue trajectory steep enough to justify doubling the price tag in under a year. If the round closes, each of the three founders would be worth roughly $4.3 billion on paper.
Kit's Honest Take: Does This Boom Actually Matter to You?
Most people will never directly touch a platform like Mercor, but it explains something real: when ChatGPT or Claude gets noticeably more accurate and hallucinates less, a meaningful chunk of that improvement traces back to invisible human labelers behind the scenes.
- The upside: better domain-expert training data means AI tools in medicine, law, and code should keep getting more reliable in specialized use cases.
- The caution: data-labeling unicorns riding this valuation wave carry the same bubble risk as AI chip stocks — revenue here is highly sensitive to any slowdown in model training demand.
- The ethics: gig-economy-style labeling work raises the same pay and working-condition questions that have long dogged this industry, and Mercor is no exception.
Rather than chasing funding-round headlines, the better signal is whether the AI tools you actually use are getting measurably more accurate in your own domain — that's the real evidence that data investment is translating into product quality.
FAQ
Q: How is Mercor different from Scale AI?
A: Both are AI data-labeling platforms. Scale AI took a major investment and leadership deal from Meta, while Mercor has stayed independent, serving OpenAI, Anthropic, Nvidia, and other labs.
Q: Is the $20B valuation confirmed?
A: Not yet — talks are ongoing as of this writing, and the $20B figure comes from sources cited in media reports, not a closed deal.
Q: Can individual developers use Mercor's services?
A: No, Mercor primarily serves large AI labs and enterprise clients rather than offering a self-serve tool for individual developers.
好不好用，試了才知道 — good or not, you won't know until you try it.
Sources / 資料來源
- Forbes: AI Data Labeler Mercor In Talks To Raise $500 Million At $20 Billion Valuation
- MarketScreener: Nvidia Mulls Mercor Investment at $20 Billion Valuation
- CNBC: AI hiring startup Mercor now valued at $10 billion with new $350 million funding round
常見問題 FAQ
Mercor跟Scale AI有什麼不同？
兩者都是AI資料標注平台，Scale AI已被Meta大舉投資並挖角高層，Mercor則走獨立募資路線，客戶涵蓋OpenAI、Anthropic、Nvidia等多家實驗室。
Mercor這輪估值200億美元確定了嗎？
截至目前仍在洽談階段，尚未正式收官，200億美元是媒體引述消息來源的預估數字。
一般開發者可以用Mercor的服務嗎？
Mercor主要服務對象是大型AI實驗室與企業客戶，並非面向一般開發者的自助工具。
延伸閱讀 / Related Articles
- AI戰鬥機評測：DARPA讓AI駕駛F-16真實試飛 | AI Fighter Jet Review: DARPA Flies F-16 Under AI Control
- Muse Code評測：Meta程式碼CLI的預設陷阱 | Muse Code Review: Meta CLI's Default Data Trap
- Etched AI晶片評測：估值一個月狂飆到210億美元硬剛Nvidia | Etched AI Chip Review: $21B Valuation Doubles, Takes on Nvidia
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言