Meta Bets Over $115B on AI Infrastructure

This spending is part of a broader vision to build "personal superintelligence" for everyone, a goal CEO Mark Zuckerberg has tied to the necessity of full general intelligence. To support this, Meta plans to have about 600,000 NVIDIA H100-equivalent GPUs by the end of 2024 alone. The strategy is a direct pivot from the company's past struggles, including a period where its AI research division, FAIR, was seen as "dying a slow death" and top talent departed for competitors like Mistral AI and Elon Musk's xAI. The massive capital outlay aims to rebuild this technical bench and regain a leadership position. This investment will materialize in several hyperscale data centers across the U.S., including the "Prometheus Hyperscale" in Ohio and "Hyperion" in Louisiana, which could eventually scale to five gigawatts. These facilities are crucial for training next-generation Llama models and supporting a future where AI is integrated into devices like Ray-Ban Meta smart glasses. Meta's open-source approach with models like Llama 2 is a key differentiator, intended to create a collaborative ecosystem and establish industry standards. Zuckerberg has stated this approach will continue, with the goal of making artificial general intelligence (AGI) widely available, a prospect that has alarmed some experts who fear the risks of uncontrolled, human-level AI. The competitive pressure is immense, with Google planning to invest up to $185 billion in its own AI infrastructure in 2026. Meanwhile, Anthropic, now a leader in enterprise AI spending, is also committing tens of billions, backed by major players like Google and Amazon, highlighting the capital-intensive nature of the current AI arms race.