Mark Zuckerberg Overhauls Meta’s Artificial Intelligence Division Amid Internal Turmoil
By Mike Isaac and Eli Tan, The New York Times, August 19, 2025
Meta CEO Mark Zuckerberg has initiated another significant restructuring of the company’s artificial intelligence (A.I.) division, aiming to accelerate its progress in the rapidly evolving and fiercely competitive A.I. landscape. The move, internally announced on Tuesday, reflects mounting internal tensions and Meta’s determination to remain at the forefront of A.I. innovation.
Reorganization of Meta’s AI Division
The latest overhaul involves splitting the existing Meta Superintelligence Labs into four specialized groups, according to sources familiar with the plans. The new structure will include:
- An A.I. research team dedicated to fundamental advancements.
- A superintelligence team focused on developing a powerful A.I. that surpasses human brain capabilities.
- A product division responsible for building and integrating A.I. features into Meta’s offerings.
- An infrastructure group overseeing data centers, hardware, and other technological resources supporting A.I. efforts.
Insiders believe that this reorganization is designed to enhance Meta’s ability to swiftly develop cutting-edge A.I. products and compete more effectively with other major players in the field. While the company aims to stabilize its structure for the foreseeable future, some A.I. executives are expected to leave amid the transition. Meta is also contemplating downsizing parts of the A.I. division, which has ballooned to thousands of employees, either by cutting roles or relocating staff to other sectors within the organization. However, final decisions on downsizing remain unsettled.
Exploring Third-Party AI Models
In a notable strategic shift, Meta is actively considering supplementing its internally developed A.I. technology with third-party models. This exploration includes potentially leveraging “open-source” A.I. systems—freely accessible for development by external parties—as well as licensing proprietary “closed-source” models from other companies. This marks a departure from Meta’s traditional reliance solely on in-house technology for powering its A.I. products.
The Push for Superintelligence and Talent Battles
The company’s intensified focus on superintelligence intensified earlier this year following challenges with advancing its flagship A.I. models. In June, Meta launched a dedicated superintelligence lab aimed at achieving an A.I. system capable of outperforming the human brain. Alongside this, Meta made a major investment of $14.3 billion in the start-up Scale AI and recruited its CEO, Alexandr Wang, as Meta’s new chief A.I. officer. The company also offered lucrative nine-figure compensation packages to attract top researchers from rivals like OpenAI and Google, fueling a competitive talent war in Silicon Valley.
During a recent investor call, Zuckerberg expressed strong confidence in superintelligence, describing it as a pathway to "a new era of individual empowerment." He highlighted that A.I. has already enhanced Meta’s core advertising business and revealed that capital expenditures could reach $72 billion this year, mostly allocated to building new data centers and expanding the A.I. workforce.
Internal Challenges and Philosophical Shifts
Since forming the superintelligence team under Wang’s leadership, internal tensions have surfaced. Wang’s team focuses on developing what they term a “frontier model” — Meta’s most powerful A.I. system yet. Unlike Meta’s historical commitment to open-source development, this team has proposed making their next A.I. model “closed,” keeping its technology proprietary and secret. This pivot has caused notable shifts within the division’s culture and operations.
Additionally, the team decided to scrap an earlier frontier model called Behemoth after disappointing performance during testing, opting to design a new model from scratch.
Talent Transitions and Changes in Leadership
The massive influx of new A.I. talent has stirred unease among some veterans in Meta’s existing A.I. ranks. In July, Meta appointed Shengjia Zhao—formerly an OpenAI researcher and co-creator of ChatGPT—as chief A.I. scientist. Zhao has been actively interviewing Meta’s established A.I. staff to evaluate their work and fit for new roles within the reorganized team.
Prominent leaders such as Nat Friedman, ex-CEO of GitHub, and Daniel Gross, founder of start-up Safe Superintelligence, are now leading the development of new A.I. features within the products and applied research division.
However, the company has seen several high-profile departures in recent months. Joelle Pineau, a foundational Meta computer scientist, left to join the start-up Cohere. Angela Fan, who contributed to Meta’s open-source Llama model, moved to OpenAI. Loredana Crisan, Meta’s vice president of generative A.I., is set to join software firm Figma as chief design officer.
Nonetheless, some longtime leaders remain. Rob Fergus, co-founder of Meta’s A.I. research division established in 2014, will continue to head the Fundamental AI Research lab (FAIR), responsible for advancing open-source A.I. research. Others like Ahmad Al-Dahle and Amir Frenkel report directly to Wang and focus on strategic generative A.I. initiatives.
Looking Ahead
Meta’s ongoing A.I. restructuring under Zuckerberg’s leadership underscores the company’s intense commitment to compete in the high-stakes field of artificial intelligence. The reorganization aims to streamline innovation around superintelligence and commercially viable products, though it brings risks of internal disruption and cultural shifts.
As Meta doubles down on building next-generation A.I., the industry will be watching closely to see if these moves position the company for success in a market increasingly dominated by a handful of major players.
For continued updates on Meta’s A.I. endeavors and the evolving tech landscape, stay tuned to The New York Times.
Correction: An earlier version misspelled the name of Amir Frenkel, one of the leaders reporting to Alexandr Wang on strategic A.I. initiatives.
Mike Isaac is a technology correspondent based in San Francisco, covering Silicon Valley and Facebook. Eli Tan reports on the technology industry from San Francisco.
