Evercore ISI analyst Mark Mahaney raised Meta Platforms Inc.'s price target to $860 from $820, implying roughly 52 percent upside from the stock's $565 close.
"In a market where almost every scaled operator is sold out, Meta is a very rare potential merchant seller of surplus capacity," Mahaney wrote in a note dated Monday.
Mahaney estimates that commercializing 0.5 to 1 gigawatt of Meta's planned 14 GW compute capacity by 2027 could generate $11 billion to $22 billion in annual gross revenue. Leasing 1 GW — just 7 percent of planned capacity — could add as much as $4.32 to annual earnings per share.
Meta plans to spend $130 billion to $145 billion this year on AI infrastructure, raising concerns about overbuilding. Renting surplus capacity would help offset that cost, though CEO Mark Zuckerberg has said selling AI services directly carries higher margins than renting compute.
Meta reportedly plans to double its computing capacity from 7 GW in 2026 to around 14 GW by 2027. One gigawatt is enough electricity to power roughly 800,000 homes.
Mahaney described the compute opportunity as a "call option" rather than a forecast, noting Meta may need all its capacity for its own AI ambitions. He does not expect Meta to become a full-scale cloud provider like Nebius Group N.V. or CoreWeave Inc.
Meta remains a massive compute customer itself, with roughly $35 billion in announced commitments to CoreWeave and up to $27 billion pledged to Nebius, including $15 billion of potential additional capacity. The company can buy capacity where it faces shortages while renting out capacity it does not need.
Zuckerberg said during Meta's latest earnings call that the company has received offers to purchase computing capacity at a premium. However, he cautioned that "it would be foolish to basically just sell all of the compute and take a short-term profit," because "there will continue to be a significantly higher margin on selling intelligence rather than selling compute directly."
Kalshi's prediction market asks what it will cost to rent an Nvidia Corp. H100 chip for one hour in July 2027. Traders gave a 60 percent chance that the average price remains above $2.50 per hour, suggesting GPU rental prices could stay strong enough to support Meta's compute opportunity.
Meta trades at about 17 times analysts' earnings estimates for the next 12 months, within 10 percent of its trailing three-year trough valuation multiple, according to Mahaney. The stock is down about 12 percent this year even as the S&P 500 has gained 12 percent.
The price target raise reflects Mahaney's view that Meta can demonstrate returns from its substantial AI investments. Investors will watch Meta's next earnings report for updates on compute utilization and any formal plans for a Meta Compute division.
This article is for informational purposes only and does not constitute investment advice.