Signs of Tension in Meta’s Partnership with Scale AI

Since June, Meta has put $14.3 billion into Scale AI, appointing CEO Alexandr Wang along with other key personnel to lead the Meta Superintelligence Labs (MSL). However, initial indications suggest that the partnership between the two organizations might face challenges.

Ruben Mayer, the former Senior Vice President of GenAI Product and Operations at Scale AI, was one of the executives brought on by Wang for MSL. Notably, he left Meta within just two months, as multiple sources revealed to TechCrunch.

Mayer came to Meta with nearly five years of experience from two stints at Scale AI. In his role, he managed AI data operations teams and reported directly to Wang, though he was not affiliated with TBD Labs, the primary initiative focused on AI superintelligence, where several prominent researchers from OpenAI have transitioned.

Mayer did not respond to numerous inquiries from TechCrunch.

Furthermore, TBD Labs is reportedly collaborating with various external data vendors alongside Scale AI to develop future AI models, according to several insiders. Among these vendors are competitors like Mercor and Surge, sources indicate.

While AI labs usually partner with several data vendors—Meta has worked with Mercor and Surge prior to launching TBD Labs—the heavy reliance on a single vendor is relatively uncommon. This raises concerns: sources report that researchers at TBD Labs view Scale AI’s data as inadequate, preferring alternatives from Surge and Mercor.

Initially, Scale AI relied on a crowdsourcing model that tapped into a large, low-cost labor pool for basic data annotation. However, as AI models have evolved, the need for highly skilled professionals—like doctors and scientists—to create and refine top-quality data has become crucial for enhanced performance.

Although Scale AI is actively working to draw in these experts through its Outlier platform, rivals like Surge and Mercor have quickly expanded, having built their business models around attracting high-paid talent from the outset.

A spokesperson from Meta downplayed concerns regarding the quality of Scale AI’s offerings. Surge and Mercor opted not to comment. When asked about Meta’s dependence on competing data providers, a Scale AI representative referenced the initial investment announcement, which highlighted a growing commercial partnership.

Meta’s collaborations with multiple third-party data vendors illustrate a diversification of its data sources, even while it invests heavily in Scale AI. Conversely, Scale AI has encountered challenges: shortly after Meta’s significant investment, OpenAI and Google announced plans to sever ties with the data provider.

Following the loss of these clients, Scale AI laid off 200 employees from its data labeling division in July, citing “shifts in market demand,” according to the new CEO Jason Droege. He noted that Scale AI intends to hire in other areas, particularly in government sales, following a $99 million contract secured with the U.S. Army.

Early speculation suggested that Meta’s investment in Scale AI aimed to attract Wang, a founder with extensive AI knowledge since Scale AI’s establishment in 2016, to enhance Meta’s recruitment of top-tier AI talent.

Beyond Wang, the overall significance of Scale AI to Meta remains uncertain.

An MSL employee mentioned that several Scale executives hired by Meta are not part of the core TBD Labs team, similar to Mayer. Additionally, it seems that Meta is not fully relying on Scale AI for its data labeling functions.

Meanwhile, the atmosphere within Meta’s AI division has reportedly become more tumultuous since Wang’s arrival and the influx of numerous top researchers, as indicated by two former employees and one current MSL staff member. Newly hired talent from OpenAI and Scale AI has voiced concerns over the bureaucracy characteristic of larger organizations, and it appears that the scope of Meta’s earlier GenAI team has diminished, they observed.

These tensions imply that Meta’s largest AI investment to date is accompanied by its own set of difficulties, despite its ambition to tackle the organization’s AI development challenges. After the lukewarm launch of Llama 4 in April, Meta CEO Mark Zuckerberg reportedly expressed disappointment regarding the AI team’s progress, as one current employee and one former employee relayed to TechCrunch.

In response to this situation and to maintain competitiveness with companies like OpenAI and Google, Zuckerberg has taken decisive steps to forge partnerships and has implemented an aggressive recruitment strategy for elite AI talent.

In addition to Wang, Zuckerberg has successfully enlisted notable AI researchers from OpenAI, Google DeepMind, and Anthropic. Meta has also acquired AI voice startups such as Play AI and WaveForms AI, alongside partnering with Midjourney, a company specializing in AI image generation.

To propel its AI objectives, Meta has revealed ambitious plans for vast data center expansions across the U.S. Among these is a $50 billion data center in Louisiana, dubbed Hyperion after a titan in Greek mythology recognized as the father of the Sun God.

Wang, who lacks a formal research background in AI, has been regarded as an unconventional choice to lead an AI lab. Reports suggest that Zuckerberg considered appointing more traditional candidates for the role, including OpenAI’s chief research officer, Mark Chen, and sought to acquire startups established by Ilya Sutskever and Mira Murati—none of whom accepted the offers.

Recently, reports from Wired indicate that several AI researchers recruited from OpenAI have already left Meta. Furthermore, numerous long-standing members of Meta’s GenAI unit have exited amid recent organizational shifts.

Rishabh Agarwal, a researcher at MSL, recently announced his departure, stating on X this week that he will be leaving the organization.

“The pitch from Mark and @alexandr_wang to build in the Superintelligence team was extremely compelling,” Agarwal remarked. “However, I ultimately chose to follow Mark’s advice: ‘In a world that’s changing so fast, the biggest risk you can take is not taking any risk.’”

When asked about his experiences at Meta and his reasons for leaving, Agarwal declined to comment.

Recently, Chaya Nayak, Director of Product Management for Generative AI, and Rohan Varma, Research Engineer, also announced their exits from Meta. The pressing question now is whether Meta can stabilize its AI operations and retain essential talent for future success.

MSL is currently focused on developing its next-generation AI model, with reports from Business Insider suggesting an expected launch by the end of the year.