Meta launches AI cloud service to sell excess compute and model access

Meta Platforms is preparing to market its surplus artificial‑intelligence computing capacity to external customers. The initiative, called “Meta Compute,” would let developers access the company’s own AI models, such as the Muse Spark model, via API, or rent raw processing power similar to services offered by Amazon Web Services, Microsoft Azure and Google Cloud. Company leaders—including infrastructure head Santosh Janardan, AI‑lab director Daniel Gross and president Dina Powell‑McCormick—have outlined two approaches: a platform‑as‑a‑service model for AI models and an infrastructure‑as‑a‑service model for raw compute, mirroring the strategies of firms like CoreWeave.

Meta’s move follows heavy investment in data‑center infrastructure and high‑end GPUs, with capital spending projected at roughly $145 billion this year. CEO Mark Zuckerberg has said that selling excess compute is “clearly an option.” The news sent Meta’s shares up about 9 % while semiconductor makers such as Micron, Intel and Nvidia saw their stocks fall. The venture makes Meta the only major hyper‑scale AI data‑center operator not yet offering cloud services, echoing recent leasing activity by Elon Musk’s xAI.

Analysts view the plan as a way to monetize the massive AI‑related capital outlays and to position Meta as a competitor in the rapidly growing AI‑cloud market.