Meta's Alexandr Wang says building an AI model is a huge blind bet

Alexandr Wang, who is chief AI officer at Meta, said building a top AI system means spending huge sums before you know it works. He said so on the podcast Cheeky Pint, hosted by John Collison, in an episode posted Oct. 12.

For most people this is about Meta's own money, and the apps it runs, like Facebook and Instagram. A bet this size shapes what the company can spend elsewhere. If you hold Meta shares in a retirement fund, some of your savings ride on it too.

Wang said Meta's AI model, Muse Wonder, had to be planned a year ahead with no way to test it along the way. Meta has put money into data centers, which are buildings full of computers, for years. He said waiting to check would leave you far behind. Then he gave his view of the startup playbook.

"It's like the opposite of lean startup." He says that begins at 52:52.

Wang gave rough math for one gigawatt of computing power, a very large amount of electricity. He said running it for a year costs about $15 billion. Building it costs more like $50 or $60 billion. At the end of last year, he said, nothing yet showed the model would work. A colleague named Susan warned him how bad failure would be financially.

Yahoo Finance reported on how the 29-year-old Wang won Mark Zuckerberg's trust and a $100 billion-plus AI bet. The story draws on the same podcast conversation.

Meta has not said whether the bet paid off, and Wang didn't give a final total.

Watch the 4 minutes where he describes the bet.51:35