Meta (META.US) is considering renting AI computing power. Zuckerberg: Resources are tight but renting at a high price can improve asset returns.

Meta Platforms (META.US) CEO Mark Zuckerberg stated that, amidst the continuous surge in demand for artificial intelligence (AI) computing power, the company still needs as much computing resources as possible. However, if external leasing offers are attractive enough, Meta does not rule out the possibility of renting out some of its AI infrastructure to enhance asset return. Zuckerberg recently stated in an interview with the media that the entire AI industry is facing a shortage of computing resources. No company believes that they have "too much" computing resources. Meta's existing computing power is almost entirely used for internal AI research and product training. However, he mentioned that if the market offers a high enough leasing price, the company will evaluate whether to rent out some computing resources to external clients rather than using them solely for internal projects. He said, "If the external offers for computing power are high enough, in some cases, renting out computing power might be more valuable than using it all internally." Earlier this month, media reports stated that Meta is studying the introduction of cloud computing services, aiming to transform the company's vast data centers and AI infrastructure into new sources of revenue. In response, Zuckerberg stated that Meta has the capability to build cloud computing services at any time, but this does not mean that the company has an excess of computing resources. It is understood that Meta is evaluating various cloud service models. One proposed solution is similar to Amazon.com, Inc.'s cloud service (AWS) Bedrock platform, which provides developers with access to AI models hosted on Meta's infrastructure, with Meta running data centers and AI chips and charging customers based on usage. The company has announced the opening of the new generation large language model Muse Spark 1.1 to developers via API, making it an important attempt to commercialize AI computing power. Regarding the hosting of models from other AI companies, Zuckerberg did not confirm, but mentioned that it is a direction worth considering. Furthermore, reports suggest that Meta also plans to introduce "bare computing power" leasing services similar to AI cloud computing companies like CoreWeave (CRWV.US), directly selling computing resources to customers. Zuckerberg mentioned that even if Meta does not fully utilize all of its computing resources in the future, the company can rent out computing resources to external clients in the long term, similar to AWS, Microsoft Corporation's Azure, and Alphabet Inc. Class C's cloud services. He specifically mentioned his interest in SpaceX's recent business model. Previously, SpaceX rented out some of its large AI data center computing power in Memphis, USA to Anthropic and reached a cooperation agreement with Alphabet Inc. Class C. According to research institutions, this model is expected to drive SpaceX's AI business to generate over $50 billion in revenue by 2028 and surpass $100 billion by 2030. Zuckerberg expressed that SpaceX's model of short-term high premium renting of computing power is "very interesting," and Meta has received similar cooperation invitations. The company will assess whether to proceed with such collaborations based on commercial value. However, he emphasized that Meta's internal demand for computing power is still very high, and the company will prioritize ensuring computing resources for its AI research and product deployment needs.