字号·· | 护眼
thenextweb

Mercor以200亿美元估值融资7.5亿美元,为各类工作训练AIMercor raises $750m at a $20bn valuation to train AI on every job

点「原文对照」整页切到原文,或双击某段只看那段的原文。

Mercor是一家旧金山初创公司,付费请专家训练和测试AI模型。该公司在D轮融资中筹集了7.5亿美元,估值达200亿美元。这一估值是Mercor在2025年10月融资3.5亿美元时达到的100亿美元的两倍。

Mercor, the San Francisco startup that pays experts to train and test AI models, has raised $750m in a Series D round that values it at $20bn. The valuation is double the $10bn Mercor reached when it raised $350m in October 2025.

Mercor在一份声明中表示,Benchmark、Felicis、Fundomo、General Catalyst、英伟达、Salesforce、Sapphire Ventures和Zoom参与了本轮融资。阿波罗旗下关联机构管理的基金,以及G Squared、Glade Brook Capital、StepStone Group和XTX Ventures管理的基金也参与了本轮融资。联合创始人布伦丹·富迪在博客文章中写道,本轮融资还包括一次员工要约收购。Mercor没有说明7.5亿美元中有多少用于此次要约收购。

Benchmark, Felicis, Fundomo, General Catalyst, Nvidia, Salesforce, Sapphire Ventures and Zoom joined the round, Mercor said in a statement. So did funds managed by affiliates of Apollo, G Squared, Glade Brook Capital, StepStone Group and XTX Ventures. The round also included an employee tender offer, co-founder Brendan Foody wrote in a blog post. Mercor did not say how much of the $750m went to the tender.

TNW在7月报道,Mercor在同意收购Deeptune时正寻求200亿美元的估值。

TNW reported in July that Mercor was seeking a $20bn valuation as it agreed to buy Deeptune.

从AI实验室到每家公司,Mercor最初是一个AI招聘平台。它现在运营着一个由500多万名专家组成的网络。这些专家用自己的专业知识训练和评估AI模型,并因此获得报酬。迄今为止,其主要买家一直是前沿AI实验室。新筹集的资金旨在把同样的服务卖给其他公司。

From AI labs to every company Mercor started as an AI hiring platform. It now runs a network of more than five million experts. They are paid to train and evaluate AI models with their professional knowledge. Its main buyers so far have been the frontier AI labs. The new money is meant to sell the same service to other companies.

“经济中的每一份工作都将需要自己的AI智能体,每家公司都需要一种在自己的工作上训练和评估这些智能体的方法,”兼任联合首席执行官的富迪说。

“Every job in the economy will need its own AI agents, and every company will need a way to train and evaluate them on its own work,” said Foody, who is co-CEO.

Mercor根据每家公司的真实工作流程构建定制测试,即所谓的评估。然后,它利用结果在该公司的工作内容上对模型进行后训练,通常从开源模型开始。今年早些时候收购的Deeptune将数百个企业应用重建为AI智能体的练习环境。

Mercor builds custom tests, known as evals, from each company’s real workflows. It then uses the results to post-train models on that company’s work, often starting from an open model. Deeptune, bought earlier this year, recreated hundreds of enterprise applications as practice environments for AI agents.

“一个模型可能在公开基准测试中得分很高,但在决定特定业务如何运行的系统、政策和边缘案例上仍然会失败,”Mercor的另一位联合创始人兼联合首席执行官阿达什·希雷马思说。

“A model can score well on a public benchmark and still fail on the systems, policies, and edge cases that define how a specific business runs,” said Adarsh Hiremath, Mercor’s other co-founder and co-CEO.

Foody援引公司内部的一个数据。他写道,Mercor在推理——即运行AI模型的成本——上的支出已经是员工薪资支出的三倍。他辩称,如果这一比例在整个经济体中保持不变,则预示着未来五年将出现接近两位数的GDP增长。

Foody gave one figure from inside the company. Mercor already spends three times as much on inference, the cost of running AI models, as it does on employee salaries, he wrote. If that ratio holds across the economy, he argued, it points to something close to double-digit GDP growth over the next five years.

根据该帖文,Mercor今年还在纽约和伦敦开设了办事处。8月,在Google赢得Spirit Airlines(美国精神航空)数据的破产拍卖后,Mercor被指定为备选买家。

Mercor also opened offices in New York and London this year, according to the post. In August it was named backup buyer for the Spirit Airlines data that Google won in a bankruptcy auction.

今年4月,Mercor发生了一起泄露事件,导致Meta冻结了与该公司的AI数据合作。Mercor于6月发布了关于该事件的最新情况说明。该声明和Foody的帖文均未提及此事。

A breach earlier this year In April, a breach at Mercor led Meta to freeze its AI data work with the company. Mercor published an update on the incident in June. Neither the statement nor Foody’s post mentions it.

阅读媒体原文 ↗