Meta’s massive $14.3B bet on Scale AI is hitting turbulence. Ruben Mayer, Scale AI’s former GenAI exec who joined Meta’s Superintelligence Labs (MSL) just two months ago, has already left. Sources told TechCrunch Mayer wasn’t on Meta’s core TBD Labs team, which builds the next-gen AI models alongside OpenAI talent.
TBD Labs is quietly sidestepping Scale AI for data. Meta’s working with rivals Mercor and Surge—both heavy hitters in AI data—despite the Scale AI investment. Researchers reportedly find Scale’s data “low quality” and favor Surge and Mercor instead.
Scale’s crowdsourcing model, relying on low-cost annotation, feels outdated as AI needs high-skill experts—doctors, lawyers, scientists—to improve training data. Competitors have leaned into this from day one, gaining ground fast. 
A Meta spokesperson pushed back on the quality claims. Surge and Mercor declined to comment. Scale AI pointed TechCrunch to its prior press on expanding the Meta deal.
Things are rough on Scale’s side too. After losing OpenAI and Google as clients post-Meta deal, Scale laid off 200 staff in July. CEO Jason Droege blamed “shifts in market demand” and is now doubling down on government contracts, including a $99M U.S. Army deal.
Meta’s AI scene looks chaotic with new hires from OpenAI and Scale frustrated by bureaucracy. Longtime GenAI staff have had their roles cut. The big bet on Scale AI, meant to fix Meta’s AI challenges after Llama 4’s flop, hasn’t smoothed out.
Founder Alexandr Wang leads MSL but isn’t a traditional AI researcher. Zuckerberg tried snagging more established AI leaders like OpenAI’s Mark Chen but failed. Some recent Meta AI recruits, including researcher Rishabh Agarwal, have already quit.
Agarwal shared his exit on X:
“The pitch from Mark and @alexandr_wang to build in the Superintelligence team was incredibly compelling,” said Agarwal.
“But I ultimately choose to follow Mark’s own advice: ‘In a world that’s changing so fast, the biggest risk you can take is not taking any risk’.”
Other recent departures include Chaya Nayak and Rohan Varma. Meta’s next AI model aims to launch by year-end but the lab’s stability and talent retention are now under real pressure.
Meta is also racing forward with massive data center builds, including a $50B Hyperion site in Louisiana. Still, the $14.3B Scale AI investment looks shakier than ever.
