Meta is exploring the sale of spare AI computing power and offering developers access to Meta-hosted models, according to Bloomberg. The social media giant, which is spending heavily on AI data centers and chips, aims to monetize that buildout by renting capacity to other companies, turning an internal supercomputer into a mini cloud service where customers would pay for computing to run AI systems and potentially use Meta’s models like Muse Spark. This would put Meta in competition with “neoclouds” such as CoreWeave and Nebius, which specialize in renting graphics processing units (GPUs). Following the report, CoreWeave and Nebius shares dropped, while Meta’s rose as traders welcomed a new way to offset its massive AI bill. If Meta has more compute than it needs at times, it can rent it out cheaply to keep data centers busy, potentially pulling down the going rate for raw GPU hours. Meta’s projected $145 billion AI spend could reprice raw GPU compute, adding a huge new supply of GPU hours and pushing pricing toward the seller’s marginal cost, which tends to squeeze neocloud unit economics first.