cracks are forming in meta’s partnership with scale ai

Meta’s Superintelligence Lab Faces Early Challenges

Since June, Meta has committed $14.3 billion to Scale AI, a data-labeling company, and brought on CEO Alexandr Wang alongside key executives to spearhead Meta Superintelligence Labs (MSL). However, indications suggest the partnership is experiencing difficulties.

Executive Departure and Role Clarifications

At least one executive who transitioned from Scale AI to MSL – Ruben Mayer, formerly the senior vice president of GenAI Product and Operations – has left Meta after only two months, according to sources familiar with the situation.

Mayer, who spent approximately five years with Scale AI, clarified his role, stating his initial focus was on lab setup rather than data operations. He also asserted his involvement with TBD Labs from the beginning and expressed satisfaction with his experience at Meta, attributing his departure to a personal matter.

Shifting Data Labeling Strategies

TBD Labs, Meta’s core AI superintelligence unit, is now collaborating with third-party data-labeling vendors beyond Scale AI to train its upcoming AI models. These include Mercor and Surge, both significant competitors to Scale AI.

While utilizing multiple vendors is common practice, Meta’s substantial investment in Scale AI makes this shift particularly noteworthy. Sources indicate researchers within TBD Labs perceive Scale AI’s data quality as lower and favor the services of Surge and Mercor.

The Evolution of Data Labeling Needs

Scale AI initially thrived on a crowdsourcing model employing a large, cost-effective workforce for basic data labeling. However, modern AI models demand highly skilled experts – doctors, lawyers, and scientists – to generate and refine high-quality data for improved performance.

Although Scale AI has introduced the Outlier platform to attract these specialists, competitors like Surge and Mercor have gained ground due to their foundational business models centered on highly compensated talent.

Meta and Scale AI Respond

A Meta spokesperson refuted claims of quality issues with Scale AI’s services. Both Surge and Mercor declined to provide comment. Scale AI directed inquiries to its initial announcement regarding Meta’s investment, emphasizing an expansion of their commercial relationship.

This development suggests Meta is diversifying its data-labeling resources, even after its significant financial commitment to Scale AI. Conversely, Scale AI has experienced setbacks, including the loss of OpenAI and Google as clients.

Scale AI’s Restructuring and New Ventures

Following the loss of key clients, Scale AI laid off 200 employees in its data labeling division in July. CEO Jason Droege attributed these changes to “shifts in market demand.” The company plans to expand into other areas, including government contracts, recently securing a $99 million deal with the U.S. Army.

Initial speculation suggested Meta’s investment in Scale AI was primarily to acquire Wang, a founder with extensive experience in the AI field since 2016, to aid in attracting top AI talent.

Questions Regarding Value and Internal Dynamics

Beyond Wang, the overall value Scale AI brings to Meta remains a subject of discussion. A current MSL employee noted that several Scale AI executives are not directly involved with the core TBD Labs team.

Furthermore, Meta’s AI unit has reportedly become more complex since Wang’s arrival and the influx of new researchers. New hires from OpenAI and Scale AI have expressed frustration with the company’s bureaucratic processes, while Meta’s existing generative AI team has seen its responsibilities curtailed.

Frustration and Recruitment Efforts

These tensions indicate a potentially challenging start for Meta’s largest AI investment, despite its aim to address the company’s AI development hurdles. Following the underwhelming launch of Llama 4 in April, Meta CEO Mark Zuckerberg reportedly expressed dissatisfaction with the AI team.

In response, Zuckerberg expedited deal-making and launched an aggressive recruitment campaign for leading AI professionals. He successfully brought on board top researchers from OpenAI, Google DeepMind, and Anthropic. Meta also acquired AI voice startups, Play AI and WaveForms AI, and formed a partnership with Midjourney, an AI image-generation company.

Infrastructure Investments

To support its AI ambitions, Meta has announced substantial data center expansions across the U.S., including a $50 billion facility in Louisiana named Hyperion.

Wang’s Role and Talent Retention

Wang, lacking a traditional AI research background, was considered an unconventional choice to lead the AI lab. Zuckerberg explored other candidates, including OpenAI’s Mark Chen, and attempted to acquire startups founded by Ilya Sutskever and Mira Murati, but all declined.

Some recently recruited AI researchers from OpenAI have already left Meta, and many long-term members of Meta’s GenAI unit have also departed.

Recent Departures and Future Outlook

MSL AI researcher Rishabh Agarwal recently announced his departure on X, praising the vision presented by Zuckerberg and Wang but ultimately choosing to “take a risk.”

Chaya Nayak, Director of Product Management for Generative AI, and Rohan Varma, a Research Engineer, have also recently announced their departures from Meta. The critical question now is whether Meta can stabilize its AI operations and retain the necessary talent for future success.

MSL is currently developing its next-generation AI model, with a targeted launch date by the end of the year.

Update: This story has been updated with comments from Mayer, who reached out to TechCrunch after publication.