Meta Has Not Disclosed a Separate Budget for Its AI Lab

Meta has not disclosed a dedicated budget for its Superintelligence Labs, leaving Mark Zuckerberg’s estimate of a possible $100 billion to $200 billion investment separate from the company’s public financial outlook.

Meta’s 2026 forecast calls for $130 billion to $145 billion in capital expenditures, including principal payments on finance leases across the company. It does not break out spending for the lab or its models. The comparable capital expenditures figure for 2025 was $72.22 billion.

Meta CEO Mark Zuckerberg described the possible $100 billion to $200 billion investment before Meta had seen the models. He characterized it as an informal estimate, not a published lab budget: “I don’t know, like $100 billion, $200 billion.”

Zuckerberg recalled a conversation with Meta Chief AI Officer Alexandr Wang that lasted about 44 minutes. Wang sent a written proposal two days later focused on a small research group and concentrated talent. Zuckerberg said Meta’s AI research group had fewer than 200 people, adding, “talent density is the thing that matters.”

Zuckerberg also gave rough infrastructure estimates: about $15 billion to operate a one-gigawatt cluster for a year, or $50 billion to $60 billion to build one. He described a cluster as just one part of model training.

Separately, Meta introduced Muse Wonder and Muse Wonder Full in mid-October. Wonder is available through Meta’s Model API and Muse Code, while Wonder Full is available to Muse Code subscribers.

Wang also recalled that Meta CFO Susan Li warned him at an executive gathering in Hawaii late last year that failure would be “very, very bad in financial terms.”