Meta explores renting AI compute to external customers

Meta's exploration of renting AI compute capacity marks a shift toward monetizing infrastructure that has been built primarily to support its own AI operations. The company's willingness to consider external customers reflects both the intensity of market demand for computing power and Meta's recognition that it may eventually face the strategic question of whether hosting third-party workloads generates more value than exclusive internal use. Zuckerberg's references to SpaceX—which is already renting its Memphis data center to AI developer Anthropic and has reached a separate agreement with Google—and to established cloud models like Amazon's Bedrock and CoreWeave's raw capacity rental suggest Meta is studying proven business structures rather than building entirely new ones.

Crucially, Zuckerberg framed this exploration as contingent: Meta currently operates at full capacity internally and sees no near-term need to rent out resources. Instead, the company appears to be designing optional revenue streams for a potential future state. The token-based pricing model for Muse Spark 1.1 through an API is already live and represents an immediate, lower-risk way to generate revenue from AI access without committing to a full cloud platform. The larger cloud offerings (hosted models and raw compute) remain under evaluation, suggesting Meta is in the assessment phase rather than launch phase. For businesses and developers, this signals that Meta may soon offer alternative ways to access AI computing resources; for investors, it suggests Meta is thinking strategically about multiple revenue sources from its infrastructure investments.