Mark Zuckerberg says Meta was ready to spend up to $200 billion on AI

American billionaire Mark Zuckerberg said Meta Platforms was prepared to spend up to $200 billion on artificial intelligence before it knew whether its new models would work, in a podcast released on Monday as the company launched its first frontier-class model.

“We’re going to write this check for like, I don’t know, $100 billion, $200 billion,” Zuckerberg said on Cheeky Pint, hosted by Stripe co-founder John Collison.

“I’m like, Alex, are you sure this is going to work?” He was speaking to Alexandr Wang, Meta’s chief AI officer, who appeared alongside him. The sum was Meta’s potential commitment, not a personal payment.

The exchange took place at the end of 2025, at a leadership offsite in Hawaii, Wang said. He had little more than a “vibe report” on the research and a few charts.

“I was feeling good about it, but it was certainly not clear at all that it was going to work,” he said. Meta’s chief financial officer, Susan Li, pulled him aside several times to press the stakes, Wang recalled.

“If it’s bad, it’s really, really, really bad,” he said she told him. “I think Susan scared you a bit,” Zuckerberg replied.

Zuckerberg said Meta was building computing capacity for running AI products before it knew the product would be ready.

“We were definitely building out this compute for inference ahead of knowing that the product was going to be ready,” he said. He argued there was room for a personal AI agent aimed at consumers, saying “no one had built the kind of personal consumer version of that.” He said the downside looked smaller than the headline sum suggested, and that he was unsure how consumers would receive Muse, Meta’s agent app.

The scale is what stands out. Wang said training a frontier model now costs “tens of billions of dollars,” and the pair put a gigawatt-scale cluster at about $50 billion to $60 billion upfront and about $15 billion a year to operate. Meta’s infrastructure has now passed five gigawatts, the podcast’s description says. The company has guided to as much as $145 billion in capital spending this year.

The models arrived on Monday. Meta released Muse Wonder, its first frontier-class model, which sits above Muse Spark, a lighter model it launched in April. Wang told Bloomberg the models are “competitive with the best models out there.” The release is positioned as a turnaround after Meta’s earlier models drew a weak reception in Silicon Valley last year. The research team behind it is under 200 people, and Zuckerberg put the research core closer to 100.

Wang runs Meta Superintelligence Labs, which reports to Zuckerberg. He joined after Meta paid $14.3 billion in June 2025 for 49% of his company Scale AI, a deal Fortune reported valued his personal stake at about $5 billion.

The bet has already moved Zuckerberg’s own wealth. Meta shares rose about 36% in September on enthusiasm for Muse, then fell on Sept. 25 and 28 after Goldman Sachs estimated that AI hyperscalers need $300 billion a year in AI revenue to break even, cutting nearly $20 billion from his fortune. Forbes put it at about $248.7 billion last week.

The test now is revenue. Meta has spent ahead of demand, and what matters is whether Muse can turn users into income at the scale of that spending.