Zuckerberg: Meta Was Ready to Commit $100-200 Billion to SI Before Knowing It Would Work
Meta CEO Mark Zuckerberg has disclosed that around the end of 2025, Meta was prepared to commit roughly $100 billion to $200 billion to super intelligence infrastructure before knowing whether its SI models would succeed.
"We're going to write this check for like, I don't know, like $100 billion or $200 billion," Zuckerberg recalled in a Financial Times interview and podcast appearance. He said he asked chief SI officer Alexandr Wang whether he was sure it would work, and Wang answered that it was "not clear at all"; CFO Susan Li repeatedly warned that the cost of failure would be enormous. Zuckerberg explained that data centers and computing capacity must be built before models can be tested commercially - the company had no option to wait for results before betting.
Last week Meta released its Muse Wonder model family, which Zuckerberg says has brought Meta to the SI frontier alongside OpenAI and Anthropic. Meta's 2026 capital expenditure guidance stands at $130 billion to $145 billion, with more than $100 billion earmarked for SI infrastructure, including its Prometheus supercluster in New Albany, Ohio. In 2025, Meta also bought a 49% stake in Scale AI for $14.3 billion and brought in co-founder Alexandr Wang as chief SI officer.
Investors have grown impatient with the burn rate: last quarter Meta reported its lowest free cash flow in years and its shares fell 8% in a day. Meta reports third-quarter earnings after the close on Wednesday, October 21, when the bill for the bet comes due for review again.
Published by GSM · USA (original Chinese edition)
Photo: 图：马克·扎克伯格（2025年）。图片来源：Wikimedia Commons，CC BY-SA 4.0，摄影：Jeff Sainlar
