Meta Platforms, Inc. is contemplating the possibility of leasing out portions of its AI computing infrastructure to external clients, provided the market offers sufficiently attractive pricing, according to CEO Mark Zuckerberg. He noted that while the demand for AI compute resources across the industry is intense and Meta currently utilizes nearly all its capacity for internal AI research and product training, the company would consider external leasing if it proves more valuable than internal use.
Zuckerberg stated in a recent media interview that no company in the AI sector feels it has an excess of computing resources. However, he indicated that if external offers for compute power are high enough, leasing it out could, in certain scenarios, be more beneficial than dedicating it entirely to internal projects.
Earlier this month, reports surfaced that Meta was studying the launch of a cloud computing business to monetize its extensive data centers and AI infrastructure. In response, Zuckerberg clarified that Meta "always has the capability to build a cloud business," but this does not imply the company has surplus compute capacity.
Meta Platforms, Inc. is evaluating various cloud service models. One approach resembles Amazon Web Services' Bedrock platform, which would provide developers with access to AI models hosted on Meta's infrastructure, with Meta managing the data centers and AI chips and charging customers based on usage. The company has already announced the availability of its new large language model, Muse Spark 1.1, via API to developers, marking a significant step in commercializing its AI compute capabilities.
When asked about potentially hosting models from other AI companies, Zuckerberg did not confirm specific plans but acknowledged it is "certainly a direction worth considering."
Additionally, reports suggest Meta is planning to offer a "bare-metal" compute leasing service similar to those provided by AI cloud computing firms like CoreWeave, directly selling computing resources to customers.
Zuckerberg expressed that even if Meta does not fully utilize all its compute resources internally in the future, the company could, like AWS, Microsoft Azure, and Google Cloud, lease computing capacity to external clients on a long-term basis.
He also mentioned a particular interest in the recent business model of SpaceX, which has leased part of the compute capacity from its large AI data center in Memphis to Anthropic and entered into a partnership with Google. Research estimates suggest this model could help SpaceX's AI business generate over $50 billion in revenue by 2028 and exceed $100 billion by 2030.
Zuckerberg described SpaceX's model of short-term, high-premium compute leasing as "very interesting" and noted that Meta has received numerous similar partnership inquiries. The company will evaluate whether to pursue such collaborations based on their commercial value.
Nevertheless, he emphasized that Meta's internal demand for compute remains very strong, and the company's current priority is to ensure sufficient computing resources for its own AI research, development, and product deployment.
Comments