Mark Zuckerberg Shakes Up Meta’s A.I. Efforts, Again

Mark Zuckerberg, the CEO of Meta, has once again reshaped his company’s artificial intelligence strategy in a dramatic reorganization that signals both ambition and internal disruption. Over the past several months, Meta has been recalibrating its approach to AI in an effort to remain a dominant player in a field increasingly defined by rapid innovation and fierce competition. This week, those efforts took a sharper turn.

According to multiple individuals familiar with the matter, Meta internally announced that its artificial intelligence division — officially known as Meta Superintelligence Labs — will be split into four separate units. Each group has been tasked with a distinct mission: one will lead cutting-edge AI research, another will focus specifically on developing superintelligence systems, a third will concentrate on consumer-facing AI products, and the last will build the infrastructure that supports these ambitions, such as advanced data centers and hardware.

A Restructuring Designed for Speed and Scale

Insiders describe this as a restructuring aimed at achieving two goals simultaneously: accelerating the development of next-generation AI models and preparing Meta to commercialize products faster in order to compete with rivals like OpenAI, Google DeepMind, and Anthropic. For Zuckerberg, the ultimate aim is to position Meta as a leader in superintelligence — AI systems that could surpass human cognitive abilities.

The reorganization is said to be a “final” structural shake-up for the foreseeable future. But while the new structure is meant to streamline decision-making, it also risks fueling additional turmoil inside the company. Several AI executives are expected to leave, while Meta is actively considering scaling down the size of the division overall. With thousands of employees already working in AI, leadership is reviewing whether certain roles should be eliminated or reassigned to other divisions. These discussions remain fluid, and final decisions have not yet been made.

Exploring External AI Models

In a significant departure from its historical approach, Meta is also weighing whether to incorporate third-party AI models into its ecosystem. This would mean moving beyond the company’s reliance on in-house technology to power AI features across Facebook, Instagram, WhatsApp, and Threads. Executives are evaluating options that include leveraging “open-source” AI models that can be freely built upon, as well as licensing “closed-source” models from other tech firms.

The move underscores Meta’s willingness to adapt in an AI race where flexibility often trumps tradition. OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude have all demonstrated how fast innovation can shift market dynamics, leaving slower-moving firms behind.

The Superintelligence Push

Meta’s renewed focus on superintelligence dates back to June, when the company announced a dedicated superintelligence lab with the audacious goal of building AI models more powerful than the human brain. At the same time, Meta made a $14.3 billion investment in Scale AI, bringing its CEO, Alexandr Wang, on board as Meta’s new chief AI officer.

To attract top-tier talent, Meta has offered compensation packages running into the hundreds of millions — igniting a poaching war across Silicon Valley. Researchers from OpenAI, Google, and Anthropic have all been targeted. In an investor call last month, Zuckerberg emphasized that he views superintelligence as “a new era of individual empowerment,” predicting that AI will transform not only consumer products but also Meta’s core business model, particularly advertising.

The company has forecasted capital expenditures of up to $72 billion this year, most of it dedicated to AI infrastructure and talent acquisition.

Tensions and Internal Culture Shifts

Since Wang took charge of the superintelligence initiative, tensions have emerged between Meta’s existing AI leadership and new arrivals. The new team is focused on developing what insiders describe as a “frontier model” — an advanced AI system meant to surpass anything Meta has built before.

This strategy includes a controversial shift away from Meta’s longstanding commitment to open-source AI. Discussions within the superintelligence team suggest that the next frontier model could be “closed,” keeping the technology proprietary. This would mark a sharp departure from Meta’s philosophy of openly releasing models such as LLaMA, which had helped establish Meta’s reputation among researchers as a champion of open AI.

The team has also scrapped Meta’s previous frontier model, called Behemoth, after poor test results and repeated delays. Instead, they are starting fresh on an entirely new system.

High-Profile Hires and Departures

Personnel changes have only added to the sense of flux. In July, Meta hired Shengjia Zhao, an OpenAI researcher best known as a co-creator of ChatGPT, to serve as its new chief AI scientist. Zhao has quickly begun reshaping the research environment at Meta, reportedly interviewing and reassessing the roles of longstanding AI staff.

Other major hires include Nat Friedman, the former GitHub CEO, and Daniel Gross, founder of Safe Superintelligence, both of whom are now tasked with leading applied AI research and product development.

But the shake-up has also triggered high-profile exits. Joelle Pineau, one of Meta’s most prominent computer scientists, left for Cohere, an AI start-up. Angela Fan, who contributed significantly to Meta’s LLaMA model, joined OpenAI. Meanwhile, Loredana Crisan, vice president of generative AI, departed to become chief design officer at Figma.

Despite this turnover, some long-serving leaders remain. Rob Fergus, who co-founded Meta’s AI research lab (FAIR) in 2014, continues to oversee day-to-day operations at FAIR, which remains focused on fundamental research and open-source contributions. Ahmad Al-Dahle and Amir Frenkel, both senior leaders in generative AI, now report directly to Wang and are steering Meta’s broader strategic AI initiatives.

Why It Matters

Meta’s restructuring comes at a time when the AI industry is consolidating around a handful of major players, each racing to build the most powerful systems. The decisions Zuckerberg makes now could determine whether Meta becomes a long-term leader or falls behind.

By investing heavily in infrastructure and talent while simultaneously entertaining external partnerships, Meta is signaling that it intends to pursue a hybrid approach: maintaining control over its most strategic AI efforts while remaining pragmatic enough to borrow from others when necessary.

For investors and industry analysts, the stakes are high. Meta’s AI strategy will shape not only the company’s future but also broader competition in the tech sector. If successful, Meta’s bet on superintelligence could redefine what consumer platforms can do — enabling smarter content recommendations, more advanced virtual assistants, and entirely new categories of products. But if the effort falters, Meta risks losing ground in an industry that punishes missteps harshly.

The Road Ahead

Zuckerberg has made clear that he is willing to restructure, reallocate resources, and even abandon past philosophies to keep Meta relevant in the AI race. While such moves create short-term uncertainty, they also reflect the urgency of a company determined not to be left behind.

The coming months will test whether Meta’s gamble pays off. With billions on the line, a growing roster of high-profile hires, and an ambitious plan to build machines more powerful than human intelligence, the company stands at a defining crossroads.

One thing is certain: the AI revolution will not wait, and neither will Zuckerberg.