Mark Zuckerberg, Meta’s CEO, is personally spearheading the creation of a 50-member elite team of AI experts to pursue artificial general intelligence (AGI), announced on June 10, 2025. The initiative aligns with a multibillion-dollar investment in Scale AI, a startup specializing in data services for AI model training. Zuckerberg has hosted recruitment meetings at his Lake Tahoe and Palo Alto homes, targeting top researchers and engineers. Internally dubbed the superintelligence group, the team will be based near Zuckerberg at Meta’s Menlo Park headquarters. The move stems from his frustration with the underperforming Llama 4 model and the delayed Behemoth model, as Meta aims to outpace rivals like OpenAI and Google. The company plans to integrate AGI into products like its chatbot, Ray-Ban smart glasses, and social platforms.
The project marks a shift to a hands-on “founder mode” for Zuckerberg, who has reorganized office spaces to accommodate the team and launched a WhatsApp group, “Recruiting Party,” to coordinate talent searches. The Scale AI investment, potentially Meta’s largest external deal, will see founder Alexandr Wang join the group post-agreement.
- Goal: Achieve AGI to lead the AI race.
- Recruitment: 50 experts, led by Zuckerberg.
- Investment: Billions in Scale AI and data infrastructure.
Meta’s strategy leverages its robust ad revenue and plans for a multi-gigawatt data center to compete globally in AI innovation.
Focus on superintelligence
Zuckerberg has made achieving artificial general intelligence a core priority for Meta, aiming to develop machines capable of human-level performance across diverse tasks. The 50-person team, including a new head of AI research, will work to embed AGI into Meta’s products, from social media platforms to AI-driven tools like chatbots and Ray-Ban smart glasses.
The initiative follows dissatisfaction with Llama 4, released in April 2025, which fell short of internal and external expectations. Developers criticized its performance, and Meta’s leadership questioned its edge over competitors like OpenAI and Anthropic. The postponement of Behemoth, touted as Meta’s most advanced model, further prompted Zuckerberg’s direct involvement.
- Objective: Integrate AGI into Meta’s ecosystem.
- Issues: Llama 4 and Behemoth underdelivered.
- Setup: Team positioned near Zuckerberg’s office.
- Applications: Social platforms, chatbots, and smart glasses.
Zuckerberg’s hands-on recruitment
Leading the recruitment effort, Zuckerberg has personally reached out to leading AI researchers, infrastructure engineers, and entrepreneurs. Over the past two months, he hosted lunches and dinners at his California residences, pitching Meta’s long-term AI vision.
In a WhatsApp group named “Recruiting Party,” Zuckerberg and senior leaders discuss potential hires around the clock. He curates his own candidate list and insists on being the first contact, emphasizing Meta’s financial strength to fund AI ambitions.
Zuckerberg highlighted that, unlike competitors reliant on large funding rounds, Meta’s ad business generates enough cash to invest tens of billions in AI, including a planned multi-gigawatt data center, one of the world’s most powerful.
Scale AI investment
Meta is nearing a multibillion-dollar investment in Scale AI, a startup offering data services for AI model training and custom applications for businesses and governments. The deal, potentially exceeding $10 billion, marks Meta’s largest external investment to date.
Scale AI’s founder, Alexandr Wang, is expected to join the superintelligence group once the agreement is finalized. The partnership aims to bolster Meta’s ability to train advanced models, a critical step toward AGI.
This investment underscores Meta’s strategy of combining internal talent with external expertise, leveraging Scale AI’s data capabilities to enhance model development.
Llama and Behemoth setbacks
Llama 4, launched in April 2025, disappointed Zuckerberg, who had pushed for a top-performing model by year-end. AI teams worked grueling schedules, but the model faced criticism for overpromising and underdelivering, both internally and among developers.
Behemoth, planned as Meta’s largest model, was delayed after leadership determined it offered insufficient advancements over prior versions. These setbacks fueled Zuckerberg’s decision to take a more active role and prioritize the new team.
- Llama 4: Criticized for poor performance.
- Behemoth: Postponed due to modest improvements.
- Response: Zuckerberg’s direct intervention.
Rivalry in the AI race
Meta is locked in fierce competition with OpenAI, Alphabet’s Google, and Anthropic to lead the AI market. Zuckerberg sees AGI as a path to dominance, enhancing Meta’s ad-driven platforms, which remain its revenue cornerstone.
The company has allocated tens of billions for AI projects in 2025, with plans for “hundreds of billions” in future years. A multi-gigawatt data center is central to this vision, providing the computational power needed to rival Google’s infrastructure.
Integration with current teams
The superintelligence group’s relationship with Meta’s existing AI teams is unclear. Some employees may shift to the new unit, but the structure is pending. The group’s creation reflects Zuckerberg’s push for breakthroughs, with the CEO closely overseeing its progress.
The team’s high profile and resources could create internal friction, as other AI divisions face pressure to perform. Its proximity to Zuckerberg underscores its priority within Meta’s strategy.
Founder-mode leadership
Over the past two months, Zuckerberg has embraced a “founder mode,” taking a more direct role in AI initiatives. He now oversees recruitment, resource allocation, and technical goals, a shift from his previously more delegated approach.
This hands-on style stems from frustration with Meta’s AI progress and a belief that the company can lead the industry. Zuckerberg’s involvement in operational details, like office layouts, highlights the strategic weight of AI for Meta’s future.
Supporting infrastructure
Meta plans to build a multi-gigawatt data center, among the world’s most powerful, to support its AI ambitions. Funded by ad revenue, this infrastructure will enable training of complex AGI models.
The Scale AI partnership ensures access to high-quality training data, addressing a key bottleneck in AI research. Together, these investments position Meta to compete at a global scale.
Market and public response
Meta’s stock remained steady on June 10, 2025, signaling investor confidence in its long-term AI strategy. Analysts welcomed the Scale AI investment and the new team, viewing Meta as a strong AI contender.
Social media buzzed with debates over AGI ethics and Meta’s growing AI influence. Users and experts discussed the feasibility of AGI and its potential integration into Meta’s consumer products.
