It’s only been since June that Meta invested $14.3 billion in vendor-scale AI that labels data, running Meta Superintelligence Labs (MSL) with CEO Alexandr Wang and some of the top startup executives. However, the relationship between the two companies already shows signs of fraying.
At least one of the kings brought in to support Run Run of Ruben Mayer, Genai Product and Operations, former senior vice president of AI. RubenMayer left Meta just two months later.
Mayer spent about five years on Scale AI across two stints. According to these sources, in his short time at Meta, Mayer oversaw the AI data operations team, but was not part of TBD Labs, the core unit within Meta that was responsible for building the AI Superintelligence, where OpenAI’s top AI researchers landed.
However, Mayer disputes some details about his role, telling TechCrunch that his first position is not data but “to help set up anything you need,” and that he is “part of the TBD Lab from day one,” rather than being removed from the core AI unit. Mayer also revealed that he “didn’t report directly to (the King)” and was “very happy” with his meta experience and left “personal issues.”
Beyond human resources changes, the relationship between meta and scale AI appears to be changing. TBD Labs is working with vendors labeled non-AI vendors to train upcoming AI models, according to five people familiar with the issue. These third-party vendors include Mercor and Surge, two of AI’s biggest competitors, people said.
AI Labs generally work with several data labeling vendors, but META has been working with Mercor and Surge even before TBD Labs spin-up, but it is rare for AI Labs to invest very large amounts in one data vendor. This makes this situation particularly noteworthy. Even Meta’s multi-billion dollar investment, several sources said TBD Labs researchers have deemed AI data poor quality and expressed their preference for working with Surge and Mercor.
Scale AI initially built its business into a crowdsourcing model that used a large, low-cost workforce to handle simple data labeling. This tag and annotate raw information to train AI models. However, the more sophisticated AI models require highly skilled domain experts, such as doctors, lawyers and scientists, to generate and refine the high-quality data needed to improve performance.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Scale AI has moved to attract experts on these subjects on an outlier platform, but competitors like Surge and Mercor have grown rapidly since they were built on the basis of well-paid talent from the start.
Meta spokesman challenged the fact that Scale AI products have high quality issues. Surge and Melkor declined to comment. Asked about Meta’s deep dependence on competitive data providers, a Scale AI spokesman cited TechCrunch as its first announcement of Meta’s investment in startups, citing the expansion of its corporate commercial relationships.
Meta’s third-party data vendors’ deals means that even after investing billions in startups, the company doesn’t put all its eggs in scale AI. However, the same cannot be said about scale AI. Shortly after Meta announced its large investment in scale AI, Openai and Google said they would halt collaboration with data providers.
Shortly after losing these customers, AI fired 200 employees in its data labeling business in July, and along with the company’s new CEO, Jason Droege, denounced some changes in “changes in market demand.” Droege said AI will serve as staff in other parts of the business, including government sales. The company said it had just signed a $99 million contract with the US Army.
Initially, I speculated that Meta’s investment in scale AI would be to seduce Wang, the founder who has been active in the AI space since AI was founded in 2016.
Apart from King, there are unresolved questions about how valuable scale AI is for the Meta.
One current MSL employee said that some of the executives on the scale, which some of the AI executives brought to Meta, are not working with the Core TBD Labs team.
Meanwhile, Meta’s AI unit has become increasingly chaotic since it brought a wave of Wang and top researchers, according to two former employees and one current MSL employee. The new talents of Openai and Scale AI have expressed dissatisfaction with navigating the bureaucracy of large corporations, but they said Meta’s previous Genai team is limited in scope.
Tensions show that Meta’s biggest AI investment to date is rocky despite being supposed to address the company’s AI development challenges. After the launch of the Llama 4 in April, Meta CEO Mark Zuckerberg was unhappy with the company’s AI team.
Zuckerberg has rushed to hit the deal, launching an aggressive campaign to hire top AI talent, in order to turn things around and catch up with Openai and Google.
Zuckerberg was able to attract Openai, Google Deepmind and top AI researchers of humanity beyond the king. Meta has also acquired AI voice startups such as Play AI and Waveforms AI, and announced a partnership with AI image generation startup Midjourney.
To promote AI ambitions, Meta recently announced the build-out of several large data centers across the United States. One of the biggest is Louisiana’s $50 billion data center called Hyperion, fathered by the sun god, named after the Greek mythological Titan.
Wang, who is not a background AI researcher, was seen as a somewhat unconventional choice to lead the AI lab. Zuckerberg has held consultations to introduce more traditional candidates, including Openai’s Chief Research Officer Mark Chen, and attempted to get Ilya Sutskever and Mira Murati startups. All of them were rejected.
Some new AI researchers brought in from Openai have already left the meta, Wired previously reported. Meanwhile, many of the long-time members of Meta’s Genai unit set out in light of the changes.
MSL AI researcher Rishabh Agarwal posted to X this week that he will be leaving the company.
“The pitch between Mark and @Alexandr_Wang for building the Superintelligence team was very appealing,” Agarwal said. “But I ultimately chose to follow Mark’s own advice: “In a world that is changing so quickly, the biggest risk you can take is not taking risks.”
When asked about his time in Meta and what prompted him to leave, Agarwal declined to comment.
Rohan Varma, head of product management at Generator AI’s Chaya Nayak and research engineer, have also announced their departure from Meta in recent weeks. The current question is whether Meta can stabilize AI operations and maintain the talent needed to succeed in the future.
MSL has already begun work on next-generation AI models. According to a report from Business Insider, it aims to launch it by the end of this year.
Update: This story was updated with comments from Mayer who reached out to TechCrunch after publication.
