Meta’s $14.3 Billion AI Gamble Shows Early Cracks, Drawing Parallels to Failed Metaverse Bet

Meta’s aggressive push into artificial intelligence, anchored by its $14.3 billion investment in data-labeling startup Scale AI, is showing signs of strain only months after the deal.

What was billed as the cornerstone of CEO Mark Zuckerberg’s campaign to position Meta at the forefront of AI superintelligence now risks becoming another costly experiment, with some analysts warning it could mirror the company’s ill-fated metaverse adventure.

According to TechCrunch, Ruben Mayer, a former Scale AI executive tapped to help establish Meta Superintelligence Labs (MSL), left the company after just two months. His departure adds to a string of talent shakeups inside MSL, where Meta has struggled to integrate high-profile recruits from OpenAI, DeepMind, and Scale AI itself.

Even more troubling, multiple people familiar with the company’s operations told TechCrunch that Meta’s own researchers are increasingly bypassing Scale AI’s data, despite the multibillion-dollar tie-up. Instead, they are turning to competitors Mercor and Surge, whose higher-cost, higher-quality annotation systems are seen as better suited to training advanced AI models.

One insider told TechCrunch that several researchers view Scale’s data as low quality, and there’s been a clear preference for other vendors.

A Fraying Partnership

Meta initially touted its investment as a way to expand its commercial relationship with Scale AI while also luring CEO Alexandr Wang and his top executives into leadership roles at MSL. The hope was that Wang’s track record and network would help Meta attract AI talent at the pace needed to catch up with OpenAI and Google.

But the relationship has shifted rapidly. After Meta’s deal, Google and OpenAI cut ties with Scale AI, and the startup soon laid off 200 workers in its labeling division. Scale’s new chief executive, Jason Droege, has tried to pivot the business toward government contracts, securing a $99 million deal with the U.S. Army.

For Meta, the pivot raises doubts about whether the enormous outlay will deliver the quality data pipeline it desperately needs. Some suggest that Meta has put billions behind a vendor that many of its own researchers don’t want to use.

Internal Chaos

The upheaval comes at a sensitive time. Zuckerberg has grown frustrated with the company’s AI output after the underwhelming release of Llama 4 in April. That disappointment set off a scramble: Meta expanded partnerships with AI voice and image startups, announced a $50 billion Hyperion data center in Louisiana, and stepped up hiring raids on OpenAI and Anthropic.

Yet the influx of new talent has bred tensions. Several recent hires from OpenAI have already quit, citing Meta’s heavy bureaucracy. Longtime staff from its generative AI unit have also departed, feeling sidelined by the new structure. Among the latest to leave was Rishabh Agarwal, a researcher who posted on X that while the vision for superintelligence was compelling, “the biggest risk you can take is not taking any risk.”

“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’.”

Director of product management for generative AI, Chaya Nayak, and engineer Rohan Varma also announced departures in recent weeks.

Déjà Vu: The Metaverse Trap

Investors and analysts are increasingly skeptical. Zuckerberg poured more than $40 billion into the metaverse, rebranding the entire company around the vision, only to see its Reality Labs unit lose more than $10 billion a year without meaningful user adoption. Now, some fear the AI gamble could follow the same trajectory.

Unlike rivals OpenAI and Google, which have already rolled out revenue-generating AI tools, Meta’s AI products remain largely in research or experimental stages. Its AI chatbot integrations across WhatsApp, Instagram, and Messenger have drawn interest but not significant revenue.

However, Zuckerberg and Wang insist MSL’s first major model will launch by year’s end. Success would allow Meta to stake a claim in the AI arms race, where Microsoft-backed OpenAI and Google’s DeepMind still lead. But if Meta continues to struggle with talent churn, vendor disputes, and questions about data quality, its $14.3 billion Scale AI bet could go down as one of Silicon Valley’s most expensive miscalculations.

For now, Wall Street remains unconvinced. Meta shares dipped last week amid reports of further AI unit departures, a reminder that investors may not grant Zuckerberg as much patience this time as they did with the metaverse.