Meta’s plan to cut approximately 600 jobs from its AI organisation looks contradictory only when hiring is mistaken for a single, steady expansion. The company spent months recruiting celebrated researchers with extraordinary compensation, paused hiring, and is now reducing established teams while protecting many of its newest arrivals.
The affected areas reportedly include FAIR research, product AI, and infrastructure groups inside Meta Superintelligence Labs. TBD Lab—the unit containing numerous high-profile hires from the summer—has been spared. After the reductions, Superintelligence Labs employs just under 3,000 people.
Meta is not spending less on AI as a whole. It is choosing which people, structures, and assets it believes can produce the breakthroughs that chief executive Mark Zuckerberg found missing from earlier efforts. That makes the layoffs a test of organisational design as much as a workforce reduction.
Llama 4 Disappointment Triggered a New Structure
Zuckerberg reportedly became dissatisfied after the lukewarm response to Meta’s Llama 4 models in April. In June 2025, the company invested $14.3 billion in Scale AI and appointed its chief executive, Alexandr Wang, as Meta’s first chief AI officer.
An aggressive recruitment campaign followed. Meta hired researchers from OpenAI including Shengjia Zhao, Jiahui Yu, Shuchao Bi, and Hongyu Ren, and reportedly recruited more than 50 people from competitors. OpenAI chief executive Sam Altman said signing bonuses reached $100 million. Former GitHub chief executive Nat Friedman and Safe Superintelligence co-founder Daniel Gross also joined.
Zuckerberg described the objective as building the industry’s most elite, talent-dense team. Yet Meta paused AI hiring in August, only weeks after the push. The sequence suggests the company had reached the point where integrating the recruits mattered more than adding another name.
Fewer People Are Expected to Make Faster Decisions
Wang told employees that the existing organisation had become too bureaucratic. A smaller team, he argued, would require fewer conversations for each decision while giving individuals broader scope, impact, and responsibility.
Legacy groups reportedly competed for computing resources, making organisational friction part of the technical problem. Protecting TBD Lab while reducing FAIR and product-oriented teams reveals a calculated preference: Meta is backing its newly assembled core over parts of the structure that preceded it.
Affected employees have been offered 16 weeks of severance plus two weeks for each completed year and encouraged to apply elsewhere inside Meta. Whether the company successfully redeploys that expertise—or releases experienced researchers to competitors—will influence the wider AI talent market.
Infrastructure Spending Shows This Is a Reallocation
One day before news of the layoffs, Meta secured a $27 billion financing arrangement with Blue Owl Capital for the Hyperion data centre in Louisiana. Cutting jobs alongside an enormous infrastructure commitment indicates a shift in resource allocation, not declining ambition.
Meta continues hiring for TBD Lab, and Zuckerberg expects the company’s AI initiatives to push expense growth in 2026 above the 2025 rate. Analysts have described the pause as a digestion period after a spending spree: a moment to decide how expensive people and compute should fit together.
The restructuring tests the assumption that concentrating exceptional researchers in a protected, startup-like unit will outperform a larger organisation with deeper institutional experience. Smaller teams can decide quickly, but they also need links to products, infrastructure, safety work, and users if research is to create business results.
Meta’s mixed headlines therefore describe one strategy. It is replacing breadth with a narrower bet on particular leaders and recruits while continuing to build the computing foundation around them. The gamble is not whether AI matters. It is whether a more concentrated organisation can turn talent and billions of dollars of infrastructure into the breakthrough that scale alone did not deliver.
