In June, Mark Zuckerberg went for the ultimate Hail Mary in the ever-intensifying AI race: He spun up a brand-new Meta AI lab after making a $14.3 billion acquisition of Scale AI — and then spending billions more hiring some of the industry’s preeminent researchers and engineers.
Fast-forward a couple of months, and Zuckerberg may have recruited the talent, but now the question is whether he’ll be able to keep it from slipping through his fingers.
For that, we’ll have to wait and see. After all, Meta rebranded its entire AI division as “Meta Superintelligence Labs,” a division of thousands of people. One team within that division — the team we’ve been hearing about the most, staffed by industry-leading AI researchers tasked with achieving superintelligence — is dubbed the “TBD Lab.” Of that team, just one active staffer, Ethan Knight, has left so far, Meta spokesperson Dave Arnold confirmed to The Verge. Knight left less than a month after joining.
Of the other two that purportedly left TBD Lab, Avi Verma and Rishabh Agarwal, they actually never started in roles with the lab, Arnold confirmed. Other staffers who have announced their departures from Meta weren’t part of the TBD Lab, but rather the full Superintelligence organization, including Rohan Varma and Meta’s director of product management for generative AI, Chaya Nayak. Nayak left for OpenAI.
Meta’s hiring spree for its Superintelligence team picked up speed months ago, with its acquisition of a 49 percent stake in Scale AI — an industry giant that provided training data to companies like OpenAI, Google, Microsoft, and Meta. As part of the deal, Scale AI CEO Alexandr Wang joined Meta to head up the lab, and Meta hired employees of OpenAI, DeepMind, Anthropic, and other companies. The most hires came from OpenAI, from which Meta reportedly poached as many as 10 staffers, shelling out some pay packages and equity that, by some reports, add up to $300 million over four years. (Meta disputes that number.)
But in recent months, AI industry staffers told The Verge that money wasn’t enough to attract certain talent, and some have turned down Meta’s reach-outs and offers, particularly for the high-visibility TBD Lab. In an industry brimming with job security and large salaries, at least for engineers and researchers of a certain level, those employees prioritize working somewhere that aligns with their values, no matter what those values may be (think: AI safety, societal impacts, accelerationism, etc.).
Last week, Meta made headlines for sending out memos about a hiring freeze and a restructuring for its AI division. The company has said both are natural follow-ups to its hiring processes in recent months. And although Meta is almost certainly tight on certain aspects of its hiring budget, it doesn’t necessarily mean the company is backtracking or changing its strategy. When a publicly-traded tech giant spends billions making new hires and wants to put all its eggs in the Superintelligence basket, it’s bound to pause and get organized — especially if it’s trying to catch up with the world’s leading labs in the ever-intensifying AI arms race. For instance, the company is reportedly no longer working on its Behemoth AI model, which failed to meet expectations internally, in favor of newer models.
So what’s going on with the hiring freeze and restructuring? A memo viewed by The Verge about the Superintelligence hiring freeze states that since Meta is beginning to plan its budget for 2026, and “with such significant change and growth in our org in 2025,” the company is “temporarily pausing hiring across all MSL [Meta Superintelligence Labs] teams, with the exception of business critical roles.”
The memo goes on to state that it’s been “common practice across many orgs across Meta in the last two years and will allow leadership to thoughtfully plan our 2026 headcount growth, as we work through our strategy.” It said that Wang’s staff will meet weekly to evaluate business-critical hires “on a case by case basis.”
Last week, Wang posted on X, “We are truly only investing more and more into Meta Superintelligence Labs as a company. Any reporting to the contrary of that is clearly mistaken.”
Another memo viewed by The Verge, posted by Wang, states that Meta’s Superintelligence division will restructure and focus on three areas: research, product, and infrastructure.
Those three areas will be covered by four teams. First up is the TBD Lab, which Wang calls “a small team focused on training and scaling large models to achieve superintelligence across pre-training, reasoning, and post-training, and explore new directions such as an omni model.”
Then there’s Meta’s Fundamental AI Research lab, or FAIR, which has been around for a while and the company has seemed to deprioritize in recent months. Its leader, Joelle Pineau, departed in May. Now, Wang wrote in the memo, FAIR will become “an innovation engine for MSL, and we will aim to integrate and scale many of the research ideas and projects from FAIR into the larger model runs conducted by TBD Lab.”
There’s also the “Products & Applied Research” team, which will “bring our product-focused research efforts closer to product development” and “include teams previously working on Assistant, Voice, Media, Trust, Embodiment and Developer pillars in AI Tech,” Wang wrote. Finally, the “MSL Infra” team will focus on “accelerating AI research and production by building advanced infrastructure, optimized GPU clusters, comprehensive environments, data infrastructure, and developer tools to support state-of-the-art research, products, and AI development across Meta,” Wang wrote. Wang also wrote that Meta would dissolve its “AGI Foundations” organization and move those employees into one of the teams above.
In a post on X, Meta spokesperson Andy Stone wrote that the hiring freeze and restructuring were “mundane” updates because “all that’s happening is: a temporary pause on some hiring while we do planning and forecasting, which we do regularly, and a structure for our new superintelligence efforts, after we brought a number of new people on board.”
Update August 29th: The article was updated to reflect more of Meta’s AI plans with specific models.
0 Comments
