In June this year, Meta Platforms, Inc. invested $14.3 billion in data annotation company Scale AI, while bringing CEO Alexandr Wang and several senior executives into Meta's Super Intelligence Laboratory (MSL). This high-profile collaboration was interpreted as a crucial step in Meta's quest to build artificial general intelligence. However, cracks began to show just two months later.

Ruben Mayer, Scale AI's former Senior Vice President who had recently joined Meta Platforms, Inc., quickly departed the company. According to two sources familiar with the matter, Mayer was responsible for the AI data operations team but was excluded from MSL's core research division, TBD Labs. However, Mayer himself refuted this claim, stating that he was a member of TBD Labs from day one, merely "helping to build the laboratory" rather than focusing on data, and did not report directly to Wang. He emphasized that his departure was "purely for personal reasons" and that he was "very satisfied" with his experience at Meta.

Regardless of how the disagreement is defined, this personnel upheaval has become the first signal of instability in the Meta-Scale AI partnership. Additionally, Meta has not entrusted all of its AI model training tasks to Scale AI. TBD Labs is currently collaborating with Mercor and Surge, both direct competitors of Scale AI. A source close to the project noted that researchers within TBD Labs believe Scale AI's data quality is subpar and prefer using annotation data provided by Mercor and Surge. This means that despite Meta's massive investment, internal trust in Scale AI has yet to be established.

**From Crowdsourcing to Professional: Why Scale AI Falls Behind**

Scale AI initially relied on a crowdsourcing model, employing large numbers of low-cost workers to handle basic annotation tasks. However, as AI models continue to evolve, data annotation is no longer simply "clicking buttons." AI systems now require extensive data support with domain expertise in medicine, law, finance, engineering, and other specialized fields – work that can no longer be entrusted to low-barrier labor platforms.

While Scale AI has attempted to attract professional talent through its Outlier platform, Surge and Mercor have been built on a foundation of "high-paid experts" from their inception, quickly gaining advantages in the refined data market.

Meanwhile, Meta's collaboration strategy is also quietly shifting. Although company spokespersons deny quality issues with Scale AI, multiple sources reveal that Meta's data quality standards far exceed previous requirements, and Scale AI's models are increasingly unable to meet demands.

More ironically, after Meta announced its investment in Scale AI, OpenAI and Alphabet both subsequently ceased their collaborations with the company. Following this, Scale AI announced layoffs of 200 employees in July, attributing this to "changing market demands." New CEO Jason Droege stated that the company would expand hiring in other business lines, particularly in government sales. This appears more like a strategic pivot: shifting from commercial AI annotation toward defense and public sector clients.

Scale AI recently won a $99 million contract with the U.S. government. However, facing distancing from three major AI giants – Meta, OpenAI, and Alphabet – Scale AI's future undoubtedly depends more heavily on government contracts for stability.

**Super Intelligence Vision Falls into Chaos: What's Meta's Next Move?**

Observers once believed that Meta's massive investment in Scale AI was truly aimed at "poaching" Alexandr Wang. This 30-year-old Scale AI founder has been active in the AI industry since 2016 and is regarded as "one of Silicon Valley's youngest AI minds." Zuckerberg clearly placed high hopes on him.

Wang's arrival triggered a series of AI talent acquisition moves at Meta, including top researchers poached from OpenAI, DeepMind, and Anthropic. Meta has also consecutively acquired voice AI companies Play AI and WaveForms AI, and announced image generation partnerships with Midjourney.

To support massive model training demands, Meta is deploying ultra-large-scale data centers across the United States. Most notably, the $50 billion "Hyperion" super center in Louisiana.

However, beyond investments, problems have not been resolved. The internal environment at MSL is becoming increasingly chaotic. Three employees (two former employees and one current employee) revealed that since Wang joined, many newcomers poached from OpenAI and Scale AI have become deeply frustrated with Meta's complex internal structures and processes.

Meanwhile, Meta's existing generative AI teams are gradually being marginalized. One researcher candidly stated, "The major incentive for joining MSL was largely the vision promises from Wang and Zuckerberg." But upon actually entering, they found the gap between vision and reality disappointing.

Some recently poached OpenAI researchers have chosen to leave, and Meta's veteran employees continue to depart. Including MSL researcher Rishabh Agarwal, Generative AI Product Director Chaya Nayak, and Research Engineer Rohan Varma, all announced their departures in recent weeks. Agarwal wrote on social media: "Zuckerberg said, 'In this rapidly changing world, not taking risks is the biggest risk.' I decided to follow his advice and choose to leave."

Meanwhile, Zuckerberg's super intelligence team faces a talent retention battle. Although Meta has begun developing next-generation AI models with plans to release within the year, current turbulence undoubtedly adds uncertainty to the project.

What was originally a bet intended to stabilize the situation through Scale AI has now become a catalyst for instability. More worryingly, this may only be the tip of the iceberg in Meta's AI empire challenges.

When a tech giant bets on a startup, they're often wagering not just on technology, but on people. Meta thought that by buying Scale AI, they could buy the future. But from personnel chaos and technical disagreements to core trust deficits, the cracks in this "super intelligence" gamble are becoming increasingly apparent.

The real issue isn't whether the investment is large enough, but whether beliefs are aligned. Meta and Scale AI's ill-timed "marriage" may be sounding an alarm for the industry: not all money-burning collaborations can burn their way to the future.