Meta Platforms (META) could be sitting on a massive untapped revenue stream that most investors have yet to fully appreciate, according to a leading Wall Street analyst.
Evercore ISI analyst Mark Mahaney raised his Meta price target to $860 per share from $820, citing a significant AI compute monetization opportunity the market has largely ignored.
The new price target implies more than 50% upside from current trading levels, backed by Mahaney’s long-term earnings outlook projecting robust GAAP EPS expansion into 2027 and 2028.
The central argument revolves around Meta’s rapidly expanding AI infrastructure, which the company plans to scale dramatically over the next two years.
Meta is reportedly planning to double its computing capacity from 7 gigawatts in 2026 to approximately 14 gigawatts by 2027, a buildout of extraordinary scale.
To put that in perspective, a single gigawatt of electricity is enough to power approximately 800,000 homes, making Meta’s planned expansion one of the largest infrastructure buildouts in corporate history.
Mahaney’s thesis holds that Meta may not require all of that capacity internally at all times, opening the door to renting surplus compute to outside customers for profit.
He estimates that commercializing just half a gigawatt of that surplus could generate $11 billion in annual gross revenue for the company.
At one gigawatt of external sales, representing just 7% of Meta’s planned total capacity, revenue could reach $22 billion and add as much as $4.32 in annual earnings per share.
Mahaney described Meta as a rare potential “merchant seller” of compute at a time when major cloud providers are effectively sold out of available capacity.
He framed the compute sales opportunity as a “call option,” deliberately stopping short of predicting Meta will transform itself into a full-scale cloud provider.
The key caveat is that Meta may ultimately need every available watt to fuel its own internal AI ambitions, which is why Mahaney is careful not to treat the $22 billion figure as a base-case forecast.
Meta has already moved to formalize its cloud ambitions through Meta Compute, a new unit designed to sell AI infrastructure and foundation models directly to third-party customers.
The company guided full-year 2026 capital expenditures to between $125 billion and $145 billion, raised from a prior range and attributed to “higher component pricing and additional data center costs.”
That spending puts Meta in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud in the increasingly lucrative AI infrastructure market.
Supporting the bullish compute thesis, prediction market Kalshi tracks the cost to rent an Nvidia H100 chip for one hour in July 2027, with traders assigning a 60% probability that average prices remain above $2.50 per hour.
Strong GPU rental pricing would be a critical factor in determining whether Meta’s surplus compute strategy can generate the kind of revenue Mahaney envisions.
Meta’s actual revenue from compute sales would ultimately hinge on utilization rates, operating costs, and how much capacity the company can consistently spare from its own AI workloads.
