Meta considered $200 billion for AI before validating its models

Meta has internally discussed expenditures that could reach up to $200 billion for AI, even as its new models remain untested. High capex forecasts and repeated warnings from the financial management frame this gamble against a backdrop of stock market nervousness.

Massive Forecasts and Stock Market Reaction

Meta's latest forecasts for 2026 estimate investment spending between $130 and $145 billion, including principal payments on financial leases. According to analysts at Cantor Fitzgerald, expenditures could hit $200 billion by 2027. This spending trajectory has raised concerns among investors: after the release of April's results and the upward revision of annual forecasts to $125 to $145 billion, Meta's stock fell by more than 6% in after-hours trading. The amounts committed in recent quarters had already raised questions among investors.

Internal Warnings and Risk Perception

Internally, Meta's management has emphasized the magnitude of the risk associated with these investments. Alexandr Wang, head of AI at Meta, reports that CFO Susan Li has warned him multiple times that failure would have severe financial repercussions. During an executive meeting in Hawaii, these warnings were reiterated. Wang mentioned a huge risk and specified that the financial management wanted to ensure that the consequences of a setback were well understood.

Amounts Considered Before Even Seeing the Models

Mark Zuckerberg stated that by the end of 2025, Meta plans to invest up to $200 billion in AI even before seeing the new models. He explained that by the end of last year, the team was excited about the progress, but the models had not yet been observed. Zuckerberg mentioned the possibility of signing a check between $100 and $200 billion and asked Alexandr Wang if he was sure it would come to fruition. Wang clarified that he had access to promising research advancements at that time, while emphasizing that there were no guarantees of success. In an episode of the Cheeky Pint podcast released on Monday, the two executives revisited these discussions. Zuckerberg added that for these investments to fail, AI as a whole would need to not work, and even in that case, Meta would have other productive uses for its computing capabilities.