Meta AI capacity question puts Zuckerberg’s cloud ambitions in focus

Meta AI capacity has become a central question for investors after Mark Zuckerberg said the company is weighing how much of its computing power to rent out and how much to keep for its own AI products. The choice matters because Meta is spending heavily on data centers at the same time its cash generation is under pressure.

On Meta’s second-quarter earnings call Wednesday, Zuckerberg said the company is receiving offers for compute at prices above what Meta paid. Compute means the chips, servers and data-center power used to train and run artificial intelligence models.

The comments came after Meta gave a third-quarter revenue forecast that CNBC described as weaker than expected. The company also said free cash flow, the cash left after operating costs and capital spending, dropped 90% from a year earlier because of higher capital expenditures.

Meta shares fell more than 7% in after-hours trading, according to CNBC. As of Wednesday’s close, the stock was down 11% for the year.

Why would Meta sell AI capacity?

Meta has been building large AI data centers, but unlike Alphabet, Amazon and Microsoft, it does not currently run a major cloud infrastructure business for outside customers. Selling or leasing unused capacity could turn part of that buildout into revenue while demand for AI computing remains tight.

CNBC reported earlier this month that Anthropic is in early discussions to lease computing power from Meta. Zuckerberg also said Meta could offer more than raw capacity, pointing to application programming interfaces, productivity services and AI agents the company is developing. An API is a software connection that lets outside developers use a company’s technology inside their own products.

Zuckerberg framed the decision as a trade-off between near-term monetization and future product development. He told investors that Meta must decide how much to earn from an asset today while still building tools, models and services that could matter later.

He also warned against selling too much of the company’s computing supply. Meta needs capacity for its own AI work, including new models overseen by AI chief Alexandr Wang. Earlier this month, Meta introduced Muse Spark 1.1, which Wang called the company’s strongest model yet for coding and agent-style work, at a lower price than offerings from OpenAI and Anthropic.

How much is Meta spending on AI infrastructure?

Meta raised the low end of its 2026 capital expenditure outlook by $5 billion, putting the range at $130 billion to $145 billion. Capital expenditures, often called capex, are long-term investments such as data centers, chips and other infrastructure.

Other large tech companies are also spending aggressively. Alphabet recently lifted the top end of its 2026 capex guidance to $205 billion and turned cash-flow negative for the first time, according to CNBC. Microsoft said Wednesday that its capex for the year will be about $175 billion. Amazon is scheduled to report results Thursday.

The size of those budgets has made investors more demanding about AI returns. Jefferies analyst Brent Thill told CNBC’s “Closing Bell Overtime” that investors want a clearer view of Zuckerberg’s plans for the compute business.

Meta’s challenge is also about sales execution. Zuckerberg said serving enterprise customers would require Meta to build a new capability as a company. CNBC has confirmed that Dave Brown, a former senior executive at Amazon Web Services, is set to join Meta.

Wall Street is judging the AI push against Meta’s recent history of large bets. Reality Labs, the unit behind virtual reality devices and wearables, lost $4.62 billion in the latest quarter on $431 million of revenue. CNBC noted that digital advertising still accounts for 98% of Meta’s revenue, which keeps pressure on Zuckerberg to prove AI can become more than another costly experiment.

This story draws on original reporting from CNBC.