Meta Platforms CEO Mark Zuckerberg is seeking employees for the company’s artificial intelligence ambitions. According to a Tuesday Bloomberg exclusive, he is frustrated with internal delays and underperformance in AI model development.
Zuckerberg has reportedly begun assembling an AI team, as he believes product delays have caused Meta to lag behind its competitors. In recent weeks, Zuckerberg has met privately with top AI engineers and researchers at his homes in Lake Tahoe and Palo Alto.
According to people familiar with the matter, discussions in these meetings are about recruitment for a secretive new unit internally dubbed the “superintelligence” group. The team is expected to have about 50 people, including a new head of AI research, most of whom Zuckerberg is recruiting himself.
Delays in AI model prompt executive action
Bloomberg reports that the CEO has reorganized office space at Meta’s Menlo Park campus so that the new team will work in close proximity to him. He also created a private WhatsApp group chat labeled “Recruiting Party,” where senior executives exchange candidate ideas and coordinate outreach at all hours.
The CEO is supposedly compiling personal lists of candidates and acting as the first point of contact in many cases.
Per the Wall Street Journal, Meta’s latest large language model, internally called Behemoth, is yet to be released after the tech firm experienced several missteps in its development. Initially scheduled for release in April with Meta’s first AI developer conference, Behemoth was postponed to June and has now been delayed again until at least the fall.
Sources inside Meta say the delay is due to less than expected performance improvements. Engineers are unsure if Behemoth offers enough advancement over previous models to justify a public release. In the meantime, Meta has launched two smaller-scale Llama models, but they have done little to answer the queries about the company’s dwindling multibillion-dollar AI plans.
These internal doubts, alongside a “lukewarm” reception to the Llama 4 model, have reportedly pushed Zuckerberg to personally oversee job hirings.
Meta’s original Llama model was built by FAIR, a team largely composed of PhDs and academic researchers. Of the 14 authors behind the original 2023 Llama paper, 11 have since left the company. Newer Llama models are now being developed by a separate team, which could be the reason for the delays.
Scale AI deal and infrastructure changes
Meta is also in advanced talks to make a multibillion-dollar investment in Scale AI, a data-labeling and infrastructure company that supports model training. If completed, the deal would become Meta’s largest external investment to date.
Scale AI’s founder, Alexandr Wang, could join the company’s superintelligence group following the agreement. Meta is set to allocate as much as $72 billion in capital expenditures this year, with a significant portion earmarked for its AI ambitions.
Zuckerberg has pitched some recruits with the firm’s advantage in cash flow, telling them the company does not need to raise external capital like its rivals and can fund massive infrastructure on its own. That includes plans for a multi-gigawatt data center, which he says could make Meta one of the most powerful AI server operators in the world.
Meta’s AI projects are led by its Fundamental AI Research (FAIR) group, so it is not clear how the new superintelligence team will interact with existing structures. Some employees are expected to transition over, but to an unknown degree of overlap or independence.
In an industry-wide AI hiatus, Meta’s competitors are also struggling to meet deadlines in their AI product releases. OpenAI, one of the industry’s biggest players, has postponed the release of its GPT-5 model, originally anticipated for mid-2025.
Your crypto news deserves attention - KEY Difference Wire puts you on 250+ top sites
