Meta signs new AI computing deals with Crusoe amid $600B infrastructure push

Meta signed new AI computing deals with data-center developer Crusoe, contracting capacity at facilities in Childress, Texas, and Warrenton, Missouri, per Bloomberg via Reuters. The agreements feed Meta's pledge to invest $600 billion in U.S. infrastructure over three years — a staggering commitment underpinning Mark Zuckerberg's bets on AI agents and the compute-hungry roadmap behind it.

The Crusoe deals are a tile in the week's dominant macro theme: compute-investment deals at unprecedented scale. NVIDIA locked in $25B of bonds, NVIDIA and SK Telecom announced a gigawatt-scale Korean cloud, and OpenAI burned $3.7B in a quarter — all pointing to an industry pouring capital into capacity faster than revenue can currently justify, on the conviction that agentic AI demand will catch up.

Meta's positioning is distinct: rather than selling cloud or chips, it is building owned and contracted capacity to serve its own products and its open-weight Llama/agentic ambitions, including the OpenEnv reinforcement-learning environment framework built with Hugging Face. Threads also crossed 500 million monthly users with new personalization features, expanding the surface Meta wants its agents to power.

The organizational caveat: Meta's head of product for the 'AI for Work' transformation, Dalton Smith, announced her departure this week, a reminder that talent churn shadows even the best-funded AI roadmaps. The skeptical question hanging over the whole $600B commitment is the one the Forbes datacenter analysis raised — whether efficient open models trained on cheaper silicon undercut the ROI on these mega-buildouts. Watch capex pacing against actual agent-revenue traction.