Meta to sell excess compute, like SpaceX
Meta plans to sell excess AI computing capacity to other companies, mirroring SpaceX’s model of monetizing surplus resources. The strategy aims to generate revenue from unused infrastructure while expanding Meta’s influence in the AI ecosystem.
Background
- Meta (Facebook's parent) is reportedly planning to sell excess computing capacity from its massive AI infrastructure to other companies — a move that mirrors what SpaceX did with its Starlink satellite internet business.
- SpaceX originally built Starlink to fund its Mars ambitions, selling internet service to generate revenue from its satellite network. Meta now appears to be applying the same logic: it built enormous data-center capacity for its own AI needs (e.g., training Llama models), and now wants to monetize spare compute cycles by renting them out.
- This is notable because Meta historically kept its infrastructure closed. Opening it up would put Meta in direct competition with cloud giants like Amazon (AWS), Microsoft (Azure), and Google (GCP), as well as specialist GPU-rental firms like CoreWeave.
- The backdrop: AI compute (especially Nvidia H100/B200 GPUs) is extremely expensive and scarce. Companies that can afford to build at massive scale — Meta, Google, Microsoft — are sitting on a resource others desperately need. Selling access is a way to turn a cost center into a profit center.