Meta’s $14B Scale AI Gamble Already Showing Cracks

Meta’s most ambitious artificial intelligence investment to date is already under strain. Just two months after committing $14.3 billion to data-labeling company Scale AI and appointing its founder, Alexandr Wang, to lead Meta Superintelligence Labs (MSL), cracks are beginning to show in what was meant to be a cornerstone partnership for Meta’s AI future.

The move was supposed to accelerate Meta’s catch-up race with OpenAI, Google, and Anthropic. Instead, it has highlighted the challenges of marrying corporate ambition with messy execution, from high-profile executive departures to concerns over the quality of Scale AI’s data.

Executive Departures Raise Red Flags

One of the first warning signs came with the abrupt exit of Ruben Mayer, Scale AI’s former Senior VP of GenAI Product and Operations, who left Meta after just two months. Mayer, initially brought in by Wang to help establish MSL, disputes claims he wasn’t central to the lab, stating he was “part of TBD Labs from day one.” He described his departure as a personal choice, but the timing highlights Meta’s challenges in retaining senior talent.

The instability doesn’t stop there. Several researchers lured from OpenAI have already walked away. Rishabh Agarwal, one of MSL’s early recruits, left citing Mark Zuckerberg’s own philosophy about risk-taking in volatile environments. High-profile figures such as Chaya Nayak, Director of GenAI product management, and Rohan Varma, research engineer, have also resigned.

This revolving door raises a serious question: Can Meta build cutting-edge AI when the very experts it needs are leaving the lab?

The Data Dilemma

Beyond staffing woes, the bigger strategic issue is data. Despite Meta’s multi-billion-dollar deal, insiders at TBD Labs reportedly prefer working with competitors Mercor and Surge, citing concerns over Scale AI’s data quality.

Scale AI has long relied on crowdsourced workers for labeling tasks. But today’s frontier models demand expert-verified data, medical imaging from doctors, legal reasoning from attorneys, and scientific annotations from researchers. While Scale AI has tried to pivot with its Outlier platform to attract professionals, rivals like Surge and Mercor, built from the ground up with highly paid domain experts, are gaining traction.

Meta, for its part, insists there are no quality issues. Still, the optics of turning to competitors after a $14B outlay suggest that the partnership isn’t delivering as promised.

Trouble for Scale AI

The struggles come at a delicate moment for Scale AI itself. Following Meta’s investment, OpenAI and Google cut ties, stripping Scale of two flagship customers. In July, the company laid off 200 employees in its data-labeling division, citing changing market conditions. Its new CEO, Jason Droege, has shifted focus toward government contracts, including a recent $99 million U.S. Army deal.

This pivot might secure new revenue streams, but it highlights how dependent Scale AI had become on big-tech clients — and how Meta’s investment has complicated, not stabilized, its future.

Zuckerberg’s High-Stakes AI Push

For Meta, this rocky start comes on the heels of April’s underwhelming Llama 4 launch, which reportedly frustrated CEO Mark Zuckerberg. Determined to catch up with AI frontrunners, Zuckerberg embarked on a hiring spree, poaching talent from OpenAI, DeepMind, and Anthropic, while also acquiring startups like Play AI (voice) and WaveForms AI (audio).

He even approved a $50 billion data center project in Louisiana, dubbed Hyperion, signaling the scale of Meta’s ambition. Yet critics question his decision to place Wang, a talented entrepreneur but not an AI researcher, at the helm of a lab tasked with building superintelligence.

Can Meta Steady the Ship?

Meta Superintelligence Labs is already working toward its next-gen model, aiming for release by year’s end. But progress will depend less on budgets and data centers, and more on stability, trust, and talent retention.

If Meta can’t resolve the growing unease within its AI unit and prove that Scale AI’s data pipeline is worth the $14B bet, it risks not only slowing innovation but also undermining confidence in Zuckerberg’s broader AI vision.

The stakes are enormous: in the AI arms race, even a short stumble can mean falling permanently behind.