Meta Raises 2026 AI Capex to $125B–$145B, Stock Drops 6%
What Happened
Previous guidance of $115B–$135B is now $125B–$145B. Q1 revenue grew 33% but the market punished the spend trajectory. Meta is now committing roughly a quarter of its market cap to infrastructure annually, with most of that pointed at AI training and inference clusters.
My Take
Meta's bet only works if open-source Llama-class models become the default substrate for enterprise AI — otherwise this is the largest unforced capex error in tech history. The contrarian read: Zuckerberg may be right. If foundation-model margins compress (see GPT-5.5 above), the value migrates to the application and ad-targeting layer, where Meta already wins. Investors panicking today will look short-sighted in 18 months — but only if Meta ships a credible enterprise distribution story this year. They haven't yet.
Read Original Source