UBS: Meta weighing a sale of excess AI computing capacity could help ease pressure on earnings

Meta Platforms is exploring ways to monetize parts of its artificial intelligence buildout that are not fully tied to serving its own apps, a development that could reduce “earnings per share” strain if spending on AI stays heavy, according to an analysis cited by Yahoo Finance.

The report characterizes a potential shift in how Meta could commercialize its AI infrastructure, including the possibility of selling “raw” computing capacity to external companies or offering access to AI models that run on Meta’s existing systems. The implication is straightforward: if some of the capacity can be billed externally, fixed costs of powering and operating data center hardware become less concentrated in Meta’s internal use.

UBS, as referenced in the Yahoo Finance piece, framed the idea as a possible offset to EPS compression risk. EPS compression typically refers to a situation where rapid growth in expenses, particularly operating costs related to new technology, outpaces revenue growth, narrowing the earnings contribution per share even if total revenues rise.

While the cited discussion points to monetization of AI infrastructure, it does not present a definitive timetable or detailed commercial terms. It also does not clarify whether the effort would be delivered through Meta’s existing advertising-and-ads ecosystem, through a standalone developer offering, or through enterprise partnerships.

Other market commentary summarized in external search results described the conversation as a “selling computing power” concept, but those posts also echoed a caution that capacity is not necessarily abundant in the broader industry. In other words, even if Meta can spare some capacity for customers, that does not automatically mean AI hardware availability is improving across the sector.

From a sector perspective, hyperscalers and AI platform builders face a similar tradeoff: deploying GPUs and accelerating servers to train and serve models takes significant capex and ongoing electricity, cooling, and operations costs. If demand for AI inference or model usage expands faster than the supply of compute, firms can typically monetize capacity at attractive rates. If demand lags, spare capacity can become an earnings drag.

In this context, any move by Meta to sell capacity or model access would be watched closely for indicates about pricing power and how much infrastructure is truly “excess.” Questions remain, including whether customers would receive dedicated compute or shared access, what performance guarantees might be offered, and how Meta would handle data privacy and model governance for external users.

For investors and customers, the next steps to watch are confirmation from Meta of whether it is actively planning such offerings, how they would be branded and sold, and whether UBS’s thesis translates into measurable improvements in cost discipline or revenue diversification. Until then, the idea remains a Wall Street hypothesis rather than a confirmed product launch.