Meta (META) reported second-quarter earnings after Wednesday’s market close, delivering stronger-than-expected revenue but falling short on earnings per share. The company also issued a third-quarter revenue forecast that came in slightly below Wall Street’s expectations at the midpoint, while raising the lower end of its 2026 capital spending guidance.
Shares of Meta dropped nearly 8% in after-hours trading following the announcement.
During the quarter, Meta posted earnings of $6.18 per share on $60.8 billion in revenue. Analysts had expected earnings of $7.14 per share and revenue of $60.2 billion.
The earnings shortfall was largely driven by one-time expenses, including $2.4 billion related to legal contingencies and $1.2 billion in severance costs. Excluding those charges, Meta would have exceeded analysts’ profit estimates.
For the third quarter, Meta expects revenue to range between $61 billion and $64 billion, compared with Wall Street’s consensus midpoint forecast of $63.1 billion.
Advertising remained the company’s primary growth engine, generating $59.3 billion in revenue, slightly ahead of expectations of $59.07 billion.
Meta also increased the lower end of its projected 2026 capital expenditures, raising its forecast to $135 billion to $145 billion, up from the previous range of $125 billion to $145 billion.
The company continues to invest heavily in artificial intelligence infrastructure. Earlier this week, Meta and BlackRock announced plans to develop a $14 billion, one-gigawatt data center in Texas. Under the agreement, BlackRock will own an 80% stake in the project, while Meta will hold the remaining 20%.
Meta CEO Mark Zuckerberg recently said the company is exploring the possibility of leasing excess data center capacity to outside customers. While no details have been announced, the strategy could resemble arrangements used by SpaceX, which has signed multibillion-dollar AI infrastructure agreements with Anthropic and Google. Another option would be to operate more like cloud infrastructure provider CoreWeave by renting computing hardware directly to customers.
The company also introduced its new Spark 1.1 AI model this month alongside an aggressive pricing strategy designed to compete with leading AI developers.
Meta charges developers $1.25 per million input tokens and $4.25 per million output tokens for Spark 1.1. By comparison, Anthropic’s Opus 4.8 model costs $5 per million input tokens and $25 per million output tokens.
By pricing its AI models well below many competitors, Meta is positioning itself to attract cost-conscious developers and potentially gain market share in the rapidly expanding AI industry.
