Meta’s AI division is undergoing a major shake-up. Just three months after appointing a new AI chief, CEO Mark Zuckerberg has initiated significant layoffs, hitting foundational research teams and even affecting star researchers. Is Zuckerberg sacrificing long-term potential for short-term gains, or is this a necessary restructuring to refocus the business?
A former Meta AI employee shared firsthand insights with Silicon Valley Observer about developing the Llama models—and the real story behind the recent cuts.
According to this insider, the open-source model Llama 4 has clearly fallen behind Chinese competitors like DeepSeek, triggering a strong sense of crisis at the highest levels. This is what pushed Zuckerberg to take bold action—bringing in outside leadership and elite talent to completely overhaul Meta’s AI strategy and structure.
Major Layoffs Hit Meta’s AI Teams
As competition in AI intensifies, top talent has become the most sought-after asset. While giants like Google and Microsoft are expanding AI teams while cutting elsewhere, Meta has taken a surprising approach—aggressively recruiting externally while laying off hundreds from its own AI division.
Last Wednesday, Meta announced it was cutting around 600 roles in its AI organization. The move was communicated by Alexandr Wang, Meta’s Chief AI Officer and head of the Superintelligence Lab, as part of an internal reorganization.
In a memo to staff, Wang stated: “By reducing team size, we will have fewer conversations required to make decisions, and each person will take on more responsibility with greater scope and impact.” Since joining Meta, Wang has focused on streamlining what he viewed as bloated structures.
The Untouchable Core Team
Meta’s current AI organization—the Superintelligence Lab—comprises four units: TBD Lab (model training and scaling), FAIR (fundamental AI research), Product Applications, and MSL Infra (infrastructure).
Three of these units were affected by layoffs—only TBD Lab emerged unscathed and continues to expand. This reveals Zuckerberg’s true AI priorities: streamlining other departments to better support TBD Labs, which carries core responsibility for developing Meta’s flagship models and AI products.
TBD Lab represents Meta’s elite AI force—a team personally led by Alexandr Wang that includes dozens of top performers from Meta’s existing AI teams alongside high-profile recruits from Google, OpenAI, and Apple, many commanding significantly higher compensation than their counterparts in other departments.
Outside Leadership Takes Charge
The restructuring was anticipated when Zuckerberg brought in Alexandr Wang this June. Wang came to Meta as part of the company’s $14.8 billion investment in Scale AI, which gave Meta a non-voting stake in the startup.
Zuckerberg valued Wang’s deep understanding of AI model training data, infrastructure, and efficient model development—exactly what Meta needed to optimize its large language models and generative AI capabilities.
Wang represents a different type of leadership—more commercially driven and execution-focused than pure researchers. In the AI arms race, rapidly turning research into usable products has become crucial.
Llama’s Stumble: The Root Cause
Wang’s appointment and the subsequent layoffs may be directly linked to Llama 4’s disappointing performance. While Meta’s open-source Llama series gained global acclaim after its February 2023 debut, the April release of Llama 4 failed to impress—just as Chinese models like DeepSeek were rapidly advancing.
According to a former Meta employee who worked directly on Llama development, the team had initially prioritized multimodal development to serve Meta’s diverse product ecosystem—including the metaverse, smart glasses, and social media.
However, when DeepSeek emerged with significantly superior reasoning capabilities early this year, it caused “panic” within Meta’s team. Attempting to excel in both areas proved impossible within the timeframe, leading to product confusion.
The former employee attributed these issues to leadership problems—product managers with limited AI backgrounds overseeing technical AI experts. While Zuckerberg and Chief Product Officer Chris Cox provided strong strategic direction, they couldn’t oversee every detail.
Zuckerberg apparently recognized these leadership issues but believed internal solutions were insufficient—leading him to bring in Wang as an external change agent.
Fundamental Research Pays the Price
The restructuring has particularly impacted FAIR (Fundamental AI Research), Meta’s foundational AI research group founded in December 2013 by Zuckerberg and AI pioneer Yann LeCun.
Under Wang’s leadership, FAIR has been integrated into the Superintelligence Lab, and its traditional research role is being replaced by more product-focused engineering responsibilities.
This shift suggests Zuckerberg’ current AI priorities favor accelerated model development and immediate returns over long-term basic research that might take decades to show value.
Talent Exodus Delights Competitors
Zuckerberg’s simultaneous aggressive hiring and extensive layoffs effectively delivers experienced AI researchers to competitors—including renowned Chinese AI researcher Yuantian “Yuan” Tian, FAIR’s Research Director, who was among those cut.
Tian, who earned his PhD from Carnegie Mellon University, has made significant contributions to reinforcement learning and large language models. He led development of DarkForestGo and OpenGo—chess AI that preceded AlphaGo—and pioneered attention mechanism innovations widely used in open-source GPT models.
On social media platform X, Tian posted: “The people responsible for the problems aren’t the ones being let go”—a pointed commentary on the situation.
His post immediately became a virtual job fair, with recruiters from OpenAI, xAI, and other AI startups expressing shock at Meta’s decision and eager to hire him and his affected colleagues.
Conclusion: DeepSeek-Driven Crisis
There’s no denying that Llama 4’s failure to keep pace with Chinese models like DeepSeek created a deep sense of crisis for Zuckerberg, prompting drastic organizational changes. Only time will tell whether bringing in Alexandr Wang and restructuring Meta’s AI division will prove successful—but the pressure from DeepSeek and other rising competitors continues to reshape Silicon Valley’s AI landscape.