Meta is weighing whether to sell AI compute to outside customers even as it commits roughly $10 billion to building more data centers, according to a report from AOL.com.

Meta is deciding whether it has enough spare capacity for forthcoming AI computing requirements to sell to others. At the same time they continue to expand their own data center footprint by investing billions of dollars. Meta has not disclosed further detail on how a compute-sales business would work, or which sites the new spending would fund, per the report.

AOL.com’s report poses a question: if Meta believes it has AI computing capacity to spare, why continue committing new capital, on the order of $10 billion, to build still more data centers?

What it means for operators: Capacity that looks spare on paper is only actually sellable if power draw and thermal headroom are visible in real time, site by site. A facility running near its power or cooling limits has nothing to offer. The more hyperscalers weigh renting out AI compute, the more the underlying question becomes how confidently they can measure the headroom they are proposing to sell.

Source: AOL.com