Meta's current AI push traces back to a single phone call. In April 2025, with Llama 4's reception falling flat, Mark Zuckerberg reached out to Alexandr Wang, the 29-year-old MIT dropout who cofounded Scale AI, for advice on fixing the company's AI effort. That first conversation ran 44 minutes. It eventually led to Meta taking a $14.3 billion stake for roughly 49% of Scale AI and installing Wang as Meta's Chief AI Officer, running a newly created unit called Meta Superintelligence Labs.
A small team, a very large budget
The scale of the financial commitment behind this reset is striking next to the size of the team actually doing the work. Meta Superintelligence Labs' core research group numbers somewhere between 100 and 200 people, a sliver of Meta's roughly 75,000-person workforce. Yet Meta's 2026 capital expenditure guidance sits between $130 billion and $145 billion, up sharply from about $72.2 billion in 2025, and Zuckerberg has floated an even larger range, $100 billion to $200 billion, as a rough sense of where total spending tied to this bet could land. By the end of 2025, those numbers were serious enough that Meta's finance chief, Susan Li, flew to Hawaii to walk through the financial stakes directly.
Paying up for talent
Some of that spending has gone into poaching talent rather than building products. Meta has reportedly dangled compensation packages as large as $100 million to pull researchers away from OpenAI, a sign of just how aggressively the company is trying to compress the time it would otherwise take to rebuild AI credibility organically. Wang himself, hired more for his track record building Scale AI's data-labeling business than for prior frontier-model research, is the clearest symbol of that willingness to buy capability rather than wait for it.
