Evercore flags a possible new Meta revenue stream from AI compute leasing

Meta Platforms is drawing renewed attention from investors for a potential business line tied to artificial intelligence compute, after Evercore ISI floated an aggressive revenue scenario for the company. In a prediction featured by Yahoo Finance, Evercore analyst Mark Mahaney suggested Meta could begin generating as much as $22 billion in annual gross revenue by renting out excess AI computing capacity starting in 2027.

The note frames AI infrastructure as a capacity problem and an opportunity. As demand for machine learning training and inference grows, companies often face the question of what to do with computing resources that are not fully utilized. Evercore’s hypothesis is that Meta, which operates large-scale data center systems to support its platforms, could monetize unused or surplus capacity by leasing it to third parties that need AI compute.

The same report also ties the estimate to a market reaction. Evercore indicated that the potential $22 billion annual gross revenue figure could translate into a stock move of more than 50%, according to the Yahoo Finance write-up. The implication is that if investors conclude Meta can consistently monetize compute, the market may re-rate the company’s earnings expectations relative to current assumptions.

A key detail is what is being measured. The estimate is described as annual gross revenue, not a profit measure. Gross revenue is sales before deducting costs, such as data center operations, power, networking, and depreciation, as well as costs of fulfilling customer demand. Without disclosure of expected margins or pricing terms, the economic significance of the revenue number remains uncertain.

Meta has not publicly, in the cited material, provided a corresponding timetable or quantified plan for AI compute leasing to external customers. In the absence of a company announcement in the post, the idea should be treated as an analyst scenario rather than a confirmed corporate strategy. That matters because compute leasing businesses typically require sustained customer demand, service-level commitments, and significant operational alignment.

More broadly, the push to commercialize AI infrastructure comes as firms seek scalable supply for training models and running AI workloads. Meta’s platforms already depend on internal AI systems for ranking, recommendations, and content moderation, which means the company has both the technical expertise and the hardware footprint to run large workloads. Whether that translates into a repeatable external-leasing business, however, is a separate question for investors to evaluate.

The report offers limited specifics on how compute would be packaged and sold, who would be the initial customers, and what form the contracts would take. It also does not describe how Meta’s internal compute allocation would be affected, or whether leasing would be incremental to existing capacity or would require reconfiguration of data center plans. Those unknowns are likely to determine the credibility of the revenue path from 2027 onward.

Investors watching Meta next may focus less on the headline dollar figure and more on follow-through indicates. That includes whether Meta provides additional disclosure about AI infrastructure monetization, whether it communicates pricing or capacity commitments, and whether customers or partners publicly point to deals involving third-party AI compute. Until then, the $22 billion estimate should be viewed as a scenario anchored to capacity economics, not as a disclosed company target.