It is only since June that Meta has invested $ 14.3 billion in the AI data supplier scale, bringing CEO Alexandr Wang and several of the startup senior executives to manage Meta Superintelligence Labs (MSL). However, the relationship between the two societies already shows signs of fraying.
At least one of the leaders that Wang brought to help manage MSL – the former main vice -president of AI Genai products and operations, Ruben Mayer – left Meta after only two months with the company, two people familiar with Techcrunch told Techcrunch.
Mayer spent about five years with an AI on a scale on two stays. In a short time at META, Mayer supervised the AI data operations teams and pointed out to Wang, but was not used to join the company’s TBD laboratories – the main unit responsible for building an IA in -altitude, where the best IA researchers from Openai landed.
Mayer did not respond to two separate requests for Techcrunch comments.
In addition, TBD Labs works with third -party data suppliers other than the AI scale to train its next AI models, according to five people familiar with the issue. These third -party suppliers include Mercor and Surge, two largest competitors on the scale, said people.
While the AI laboratories generally operate with several data suppliers – Meta has been working with Mercor and Surge since TBD Labs has been turned – it is rare for an AI laboratory to invest so strongly in a data supplier. This makes this situation particularly notable: even with the investment of several billion dollars of Meta, several sources said that TBD Labs researchers considered data on the AI scale as of low quality and expressed a preference to work with overvoltage and mercore.
Scale IA initially built its activities on a crowdsourcing model that used a large low-cost labor to manage simple data annotation tasks. But as AI models have become more sophisticated, they now require highly qualified domain experts – such as doctors, lawyers and scientists – to generate and refine high quality data to improve their performance.
Techcrunch event
San Francisco
|
October 27-29, 2025
Although the AI scale has decided to attract these experts in the matter with its aberrant platform, competitors like Surge and Mercor increased quickly because their commercial models were built on a well-remunerated talent base from the start.
A Meta spokesperson disputed the fact that there are quality problems with the product of the AI scale. The overvoltage and the mercore refused to comment. Asked about META’s deepening of dependence on competing data providers, a spokesperson for AI of the scale led TechCrunch to its initial announcement of Meta’s investment in the startup, which quotes an expansion of the business relationship of companies.
Meta agreements with third -party data providers probably mean that the company does not put all its eggs on the AI scale, even after having invested billions in the startup. However, the same cannot be said for the AI scale. Shortly after Meta announced her massive investment with the AI scale, Openai and Google said they would stop working with the data supplier.
Shortly after losing these customers, the AI scale dismissed 200 employees in its data labeling activity in July, with the new CEO of the company, Jason Droege, blaming the changes in part on “market demand changes”. Droege said that the AI scale is rising in other parts of the company, including government sales – the company has just won a $ 99 million contract with the US military.
Some initially supposed that Meta’s investment in the AI scale should really attract Wang, a founder who operated in AI space since the AI scale was founded in 2016 and who seems to help Meta attract the best AI talents.
Aside from Wang, there is an open question on the value of the meta scale.
A current MSL employee says that several of the scale executives brought to Meta do not work in the basic team of TBD Labs, as with Mayer. In addition, Meta is not exclusively based on scale AI for data labeling work.
Meanwhile, the Meta AI unit has become more and more chaotic since it caused Wang and a wave of better researchers, according to two former employees and an current MSL employee. New Talents of Openai and the AI scale expressed their frustration of navigating the bureaucracy of a large company, while the previous Genai team of Meta saw its limited scope, they said.
Tensions indicate that the biggest investment in Meta AI to date may have gone to a difficult start, despite the fact that it was supposed to meet the challenges of development of the company. After the dull launch of Llama 4 in April, the meta-PDG Mark Zuckerberg became frustrated by the company’s AI team, a current and a former employee told Techcrunch.
In an effort to overthrow the steam and catch up with Openai and Google, Zuckerberg rushed to conclude offers and launched an aggressive campaign to recruit the best talents of AI.
Beyond Wang, Zuckerberg managed to attract the best IA researchers from Openai, Google Deepmind and Anthropic. Meta has also acquired AI vocal startups, notably Play AI and Wave forms AI, and has announced a partnership with the AI image generation startup, Midjourney.
To fuel its AI ambitions, Meta recently announced several buildings of massive data center in the United States, one of the largest is a dollar data center in Louisiana called Hyperion, named after a Titan in Greek mythology that caused the Sun God.
Wang, who is not a background researcher, was considered a somewhat unconventional choice to lead an AI laboratory. Zuckerberg would have had talks to bring more traditional candidates to direct the effort, such as the research director of Openai, Mark Chen, and tried to acquire the startups of Ilya Sutskever and Mira Murati. All refused.
Some of the new IA researchers have recently brought Openai have already left Meta, Wired previously reported. Meanwhile, many long -standing members of Meta’s Genai unit have gone in light of changes.
MSL AI researcher, Rishabh Agarwal, is among the last, displaying this week that he would leave the business.
“The field of Mark and @alexandr_wang to build in the superintendent team was incredibly convincing,” said Agarwal. “But I finally choose to follow Mark’s own advice:” In a world that changes so quickly, the biggest risk you can take is not to take any risk “.”
Asked later on her stay in Meta and what led her decision to leave, Agarwal refused to comment.
The Director of Products of the Generative AI, Chaya Nayak, and the research engineer, Rohan Varma, have also announced their departure from Meta in recent weeks. The question is now whether Meta can stabilize its AI operations and keep the talent he needs for his future success.
MSL has already started working on its next generation AI model. According to the Business Insider reports, it aims to launch it by the end of this year.
