In a whirlwind of strategic shifts, Meta is gearing up for its fourth reorganization of its artificial intelligence (AI) divisions within just six months, signaling an aggressive push to stay competitive in the rapidly evolving AI landscape.
The company’s latest move involves restructuring its newly formed Meta Superintelligence Labs (MSL) into four distinct groups, with a focus on accelerating the pursuit of superintelligence — an ambitious goal to create AI systems that surpass human cognitive abilities. This follows a series of high-profile changes, massive talent acquisitions, and significant financial investments, all aimed at positioning Meta as a leader in the AI race.
A Tumultuous Six Months
Meta’s AI journey has been marked by frequent reshuffling. In **February 2025**, the company made headlines by relocating two engineering leaders from its AI group and appointing Loredana Crisan, then head of Messenger, to lead product development — an early indicator of the turbulence to come. By **May**, Meta split its generative AI group into two separate teams: one focused on research and another on products, aiming to streamline efforts amid growing competition from rivals like OpenAI, Google, and Anthropic.
In **June**, CEO Mark Zuckerberg doubled down on Meta’s AI ambitions, embarking on a multi-billion-dollar hiring spree to attract top talent from leading AI firms. This included a staggering $14.3 billion investment in Scale AI, which brought its founder, Alexandr Wang, to Meta as chief AI officer, alongside a team of Scale AI employees.
The creation of Meta Superintelligence Labs was announced as a bold step toward developing artificial general intelligence (AGI) and beyond. However, the rapid pace of change and internal tensions have kept Meta’s AI division in flux, culminating in the latest restructuring announced in August 2025.
Meta Superintelligence Labs: A New Structure
The newly restructured Meta Superintelligence Labs is being divided into four specialized groups, each with a distinct mandate to advance Meta’s goal of achieving superintelligence:
- TBD Lab: A new lab, curiously named “To Be Determined,” will focus on building and refining foundational AI models, including future iterations of Meta’s Llama series. This group is tasked with tackling core research challenges, such as pre-training, reasoning, and post-training, while exploring innovative directions like multimodal “omni” models capable of handling text, images, and video.
- Products Team: Led by Nat Friedman, former CEO of GitHub, this team will integrate AI into Meta’s consumer offerings, such as Facebook, Instagram, WhatsApp, and the Meta AI assistant, which already serves over one billion monthly active users.
- Infrastructure Team: Headed by Aparna Ramani, a long-time Meta engineering vice president, this group will oversee the company’s AI infrastructure, including data centers and hardware, to support the computational demands of advanced AI models.
- Fundamental AI Research (FAIR) Lab: Led by Robert Fergus, a co-founder of FAIR, this team will continue its focus on long-term AI research, contributing foundational insights to Meta’s broader AI strategy.
A Star-Studded TBD Lab Leadership
The TBD Lab, in particular, stands out for its high-profile leadership, entirely composed of external hires from top AI organizations.
These six leaders bring a wealth of expertise to Meta’s superintelligence efforts:
- Jack Rae (ex-Google): Rae, who previously led pre-training for Google DeepMind’s Gemini models and spearheaded reasoning development for Gemini 2.5, will oversee pre-training efforts, where models are trained on vast datasets to predict text and other patterns.
- Ruoming Pang (ex-Apple): Formerly head of Apple’s Foundation Models team, Pang will lead infrastructure for TBD Lab, a role distinct from the broader infrastructure group under Ramani. His experience in model development at Apple positions him to tackle the computational challenges of scaling large AI models.
- Jiahui Yu (ex-OpenAI): Yu, a co-creator of OpenAI’s GPT-4o and other advanced models, will head multimedia initiatives, enabling Meta’s models to understand and generate non-text content like images and videos.
- Hongyu Ren (ex-OpenAI) and Pei Sun (ex-Google): This duo will co-lead post-training efforts, using refined datasets to enhance model performance on specific tasks. Ren contributed to OpenAI’s GPT-4o and o-series models, while Sun worked on post-training and reasoning for Google DeepMind’s Gemini models.
- Shengjia Zhao (ex-OpenAI): Named chief scientist for MSL, Zhao, a co-creator of ChatGPT and GPT-4, will direct research efforts, reporting directly to Alexandr Wang.
This “dream team” reflects Meta’s aggressive talent acquisition strategy, with substantial compensation packages — some reportedly reaching nine figures —luring top researchers from competitors. However, the absence of Alexandr Wang’s name in the leadership structure of TBD Lab is striking, given his role as chief AI officer and the significant investment Meta made in Scale AI to secure his expertise. While Wang is overseeing MSL as a whole, his lack of mention in the TBD Lab’s leadership raises questions about his specific role in the restructured organization.
Dual Infrastructure Groups and Strategic Shifts
A notable aspect of the reorganization is the creation of two separate infrastructure groups. The TBD Lab’s infrastructure team, led by Ruoming Pang, will focus on the specific needs of building and scaling foundational models. Meanwhile, Aparna Ramani’s broader infrastructure group will manage Meta’s overarching AI infrastructure, including data centers and hardware, ensuring alignment across MSL’s efforts.
This dual structure aims to balance the specialized demands of TBD Lab with the company’s broader computational requirements, which Zuckerberg has backed with investments projected to reach $72 billion in 2025 alone.
The reorganization also involves dissolving the AGI Foundations team, created just months earlier in May, with its members redistributed across the new groups. This move, coupled with the reported abandonment of Meta’s Behemoth frontier model due to poor performance, underscores the challenges Meta faces in advancing its AI capabilities. The company is now exploring a shift from its open-source philosophy — exemplified by the Llama models — to potentially developing a closed model, a significant departure that could reshape its competitive strategy.
Internal Tensions and Industry Implications
Meta’s rapid restructuring has not been without challenges. The influx of highly compensated external hires has sparked tensions with existing AI researchers, some of whom have threatened to leave. Notable departures include Joelle Pineau, who joined Cohere, Angela Fan, who moved to OpenAI, and Loredana Crisan, who left for Figma as chief design officer. These exits, combined with the lukewarm reception of Llama 4 and delays in the Behemoth model, highlight the internal and external pressures Meta faces in its quest for superintelligence.
The reorganization also reflects Meta’s broader strategy to keep pace with rivals like OpenAI, Google DeepMind, and Anthropic, who have made significant strides in advanced AI development. By assembling a world-class team and investing heavily in infrastructure, Meta is signaling its determination not to be left behind. However, critics, including Meta’s own chief AI scientist Yann LeCun, have questioned whether scaling large language models (LLMs) alone can achieve superintelligence, suggesting that new architectural approaches are needed.
Also read:
- Google Agrees to $36 Million Fine in Australia Over Pre-Installed Search Deals
- Wall Street AI Stocks Slide Triggered by MIT Report
- Google Has Indexed Over 370,000 Grok Chats, Exposing Sensitive Data
- Tensions Rise in UK Financial Regulation as Bank of England Blocks Revolut MeetingTensions Rise in UK Financial Regulation as Bank of England Blocks Revolut Meeting
Looking Ahead
Meta’s fourth AI restructuring in six months is a bold attempt to streamline its efforts and capitalize on its massive investments in talent and infrastructure. The creation of Meta Superintelligence Labs, with its four specialized groups and an all-star leadership team, positions the company to tackle the technical and strategic challenges of building superintelligent AI. However, the absence of clarity around Alexandr Wang’s role in TBD Lab, the dissolution of recent AI units, and ongoing internal tensions raise questions about execution and stability.
As Meta navigates this high-stakes race, its ability to balance long-term research with consumer-focused product development will be critical. With billions of dollars and some of the brightest minds in AI at its disposal, Meta is betting big on superintelligence — but whether it can deliver on this ambitious vision remains to be seen.
*Sources: The Information, Bloomberg, The New York Times, Reuters, Business Insider, WIRED, CNBC
