Meta is on pace to spend somewhere between $100 billion and $200 billion on artificial intelligence, with 2026 capital expenditures guidance running as high as $145 billion. A single one-gigawatt data center costs an estimated $50 billion to $60 billion to build and roughly $15 billion a year to operate, and the company's Hyperion campus in Louisiana is planned to eventually scale to five gigawatts, giving a sense of just how large the underlying infrastructure bill has become.
Mark Zuckerberg has framed that spending as the direct result of a lesson he says he learned from Apple. Meta spent years at the mercy of decisions made by Apple and Google, including Apple blocking a Facebook Research app from its App Store in 2019 and rolling out App Tracking Transparency restrictions in 2021 that Meta has said cost it roughly $10 billion in 2022 alone. That history is what Zuckerberg points to when explaining why Meta is now building a full stack operation of its own, owning its data centers, chips, and infrastructure rather than depending on platforms it does not control.
The buildout is meant to support a wave of new AI products. Alexandr Wang, Meta's chief AI officer, now leads the company's superintelligence lab, which launched a frontier AI model called Muse Wonder on October 12. Meta is also rolling out Muse, a personal AI agent meant to handle tasks like shopping, financial planning, legal questions, and medical information, with paid subscription tiers priced at $20 and $100 a month. Separately, the company is building a cloud business designed to sell off excess AI computing capacity it is not using internally.