AI training companies are raising billions to get humans to teach chatbots. Here are the startups cashing in.

Based in San Francisco and founded in 2016, Scale AI is the best-known of the AI training startups — especially after Meta bought 49% of it for $14.3 billion this summer, recruiting its CEO, Alexandr Wang, for its superintelligence team. This triggered an exodus of customers like Google, with the startup later overseeing layoffs. Scale AI still commands one of the largest workforces in the sector, with over 300,000 gig workers across its Outlier and Remotasks platforms, and says it's pivoting to enterprise and military deals. As of July, it's unprofitable, Business Insider previously reported.

Once under the radar, San Francisco-based Surge AI has made headlines for saying it was able to hit $1.2 billion in revenue last year without raising any venture capital funding. The result? Its founder and CEO, Edwin Chen, who owns a 75% stake in the company, is now worth a whopping $18 billion, Forbes reported, pegging his company's current valuation at $24 billion. That could go even higher if the startup closes a billion-dollar fundraising round at a $30 billion valuation, per Forbes. Surge AI, which was founded in 2020, runs the gig platform Data Annotation, where it says it pays 1 million gig workers to train AI models and touts rates of over $40 an hour. A spokesperson added that some specialists can make over $200 an hour. Surge AI's customers include Anthropic.

Founded in 2023, San Francisco-based Mercor uses AI interviews to match people to specific AI training gigs. In a podcast released earlier this month, Mercor's 22-year-old CEO and cofounder, Brendan Foody, said the company pays contractors an average of $95 an hour. It is in talks to raise at a $10 billion valuation, according to a source familiar with the matter.

The popular Gen Z job search platform expanded into AI training earlier this year after its CEO, Garrett Lord, noticed frontier labs' demand for human trainers in areas like accounting, law, and medicine, and STEM. Workers on Handshake can make $75 to $175 an hour as of this fall, a spokesperson told Business Insider. The company was last valued at $3.5 billion in January 2022.

Turing focuses more on coding and technical talent than its competitors, connecting software engineers to top AI labs like Anthropic and Google's Gemini team. Based in San Francisco and founded in 2018, Turing raised $111 million at a $2.2 billion valuation, it announced in June. It tripled its annual revenue runrate to $300 million last year and is profitable, it said in a press release.

Founded in 2015, this startup was early to the AI boom, with its major claim to fame being that it helped OpenAI train the original ChatGPT. Invisible offers foundational AI companies a marketplace of human trainers they can choose from to improve their AI models, along with other enterprise AI solutions. It announced a $100 million fundraise this month, valuing it at over $2 billion. Invisible laid off dozens of contractors in 2023, Business Insider previously reported, in the months after OpenAI released ChatGPT.

Spun out of Stanford University, Snorkel AI creates datasets validated by people in fields ranging from STEM to the humanities. It touts partners like Google, Mistral, and Anthropic on its website and creates tests where AI companies can figure out whose chatbot is better. Snorkel AI is based in Silicon Valley and was founded in 2019. It raised $100 million in Series D funding at a $1.3 billion valuation in May, it announced. Snorkel cut 13% of its workforce earlier this month.

Labelbox is best-known for its gig work site Alignerr. Its CEO Manu Sharma told Forbes the company is "basically a unicorn" — meaning it's valued at around $1 billion — for its $110 million Series D back in 2022. Labelbox mentions customers like Google Cloud and Walmart on its website, while Alignerr touts rates of $20 to $120 an hour online.

Micro1 connects AI labs to human trainers through an AI interview service. Founded in 2022 and based in Silicon Valley, Micro1 announced it raised $35 million in Series A funding at a $500 million valuation this month. It has hit over $60 million in annual revenue, its founder and CEO, Ali Ansari, told Business Insider. Its customers include Microsoft and other Big Tech companies, Ansari added.

One of the oldest players in the data labelling game, Appen was founded in 1996 in Sydney to train voice recognition technologies. The company has since expanded its AI training operations to 170 countries and is publicly traded in Australia. It counts Meta, Nvidia, and Salesforce as some of its Big Tech partners and says it has amassed over 1 million contractors. The company is worth about $150 million, and its stock is down over 70% so far this year. A spokesperson told Business Insider that Appen is potentially significantly undervalued thanks to high China revenues.

GlobalLogic is a subsidiary of Japanese conglomerate Hitachi. It provides AI trainers to tech companies like Google, Wired reported, recently laying off 200 of them. Hitachi bought GlobalLogic for $9.6 billion in 2021.

Have a tip? Contact Charles Rollet via email at crollet@insider.com or on Signal and WhatsApp at 628-282-2811. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.