Meta may sell AI compute capacity to offset $125B data center costs

Meta is exploring a major shift: selling access to its AI data center capacity to third parties, aiming to defray the staggering costs of its infrastructure buildout. The company spent $72 billion on capital expenditures in 2025 (mostly AI-related) and plans to spend $125-145 billion in 2026. Options under consideration include launching a dedicated cloud division, offering raw compute to select buyers, or enabling capacity on Meta’s existing AI infrastructure like Muse Spark. This would position Meta against neocloud providers such as CoreWeave, Nscale, and Nebius—the latter of which Meta already has a $27 billion capacity deal with.

Meta’s move echoes SpaceX’s strategy of selling excess data center capacity, including a $1.25 billion/month deal with Anthropic and $920 million/month with Google. However, Meta faces an uphill battle in AI adoption: its Meta AI app trails far behind ChatGPT, Google Gemini, and Claude in usage, and its enterprise AI presence is virtually nonexistent. The plan could transform costly infrastructure into a new profit center beyond advertising, following failed metaverse attempts. If successful, it could disrupt the neocloud market and provide a secondary revenue stream for Meta’s AI ambitions.