Cracks are forming in Meta’s partnership with Scale AI

Meta’s multi-billion-dollar investment in Scale AI, designed to supercharge its AI ambitions and staff the Meta Superintelligence Labs (MSL), is facing unexpected turbulence, according to recent reports from TechCrunch. Less than three months after Meta brought in Scale AI's CEO Alexandr Wang and several executives, key figures are already leaving, and MSL is sourcing data from competitors like Mercor and Surge, despite its sizable partnership with Scale AI.

Executive Departures and Internal Tensions

Scale AI’s former Senior VP, Ruben Mayer, exited Meta after just two months, signaling potential discord beneath high-profile hires. Several other Scale AI executives at Meta are reportedly not involved with the pivotal TBD Labs team, the group leading Meta’s “AI superintelligence” push. Exits aren’t limited to Scale alumni; AI researchers lured from OpenAI and Meta’s own GenAI team have also departed, citing frustration with Meta’s bureaucratic environment and shifting team scopes.

Quality Concerns and Data Strategy Shifts

Five sources confirmed that TBD Labs is buying training datasets from Mercor and Surge, two of Scale’s largest rivals, while expressing doubts over Scale’s data quality. Scale AI’s historic focus on crowdsourced annotation is increasingly seen as outdated for the specialized needs of advanced AI, as systems now require domain experts like doctors, lawyers, and scientists for generating and refining training data. Scale’s Outlier platform aims to compete, but rivals built around highly-paid talent are gaining traction.

Market Fallout and Meta’s Next Moves

After Meta’s investment announcement, both OpenAI and Google ceased business with Scale AI, leading to substantial layoffs, 200 employees left Scale’s data labeling division in July, attributed to shifting market demand by the new CEO, Jason Droege. Droege noted Scale would expand government and enterprise sales, evidenced by a $99 million Army contract win. Meanwhile, some speculate that Meta’s real goal was attracting Wang and other top talent, as Mark Zuckerberg seeks fresh leadership and expertise amid AI setbacks, including the lukewarm reception of Llama 4 earlier this year.

The Road Ahead

Meta continues to recruit aggressively and invest, building out massive U.S. data centers like the $50 billion Hyperion facility in Louisiana. The company has absorbed voice startups and inked a deal with image generation firm Midjourney, further intensifying development for its next-gen AI model, which is slated to launch before year-end. But with ongoing leadership churn and team fragmentation, questions linger over Meta’s ability to stabilize its AI operations and actualize the transformation its investment in Scale AI was meant to deliver.