Meta's Zuckerberg Weighs Selling AI Compute Against Keeping It for Himself

Mark Zuckerberg spent part of Meta's second-quarter earnings call this week doing something unusual for him: thinking out loud about turning excess AI computing power into a business, while stopping well short of promising he'll actually do it.

The question at the center of it all is simple to state and hard to answer. Meta is spending tens of billions of dollars building AI data centers. Some of that capacity, at least for now, sits beyond what Meta's own AI models and products need. Other companies desperately want it. So does Meta sell the extra compute, rent it out, or hoard it for whatever comes next?

"We're getting a lot of offers for compute at a significant premium over what we paid for it," Zuckerberg said on the call Wednesday. "And we have more coding and productivity tools on our roadmap as well."

That's about as close as he got to a plan. But the backdrop makes clear why the question is suddenly urgent.

A cloud business Meta doesn't have

Of the four major U.S. hyperscalers, Meta is the only one without a cloud infrastructure business, even though its capital spending rivals Amazon's, Microsoft's and Google's. That's a strange position to be in during an AI boom where computing capacity is the scarcest resource in tech. Zuckerberg has floated the idea of launching a cloud offering in recent months, and CNBC reported earlier this month that Anthropic — one of the AI labs Meta competes with on the model side — is in preliminary talks to lease computing power directly from Meta.

The timing of Zuckerberg's comments wasn't great. Meta issued a weaker-than-expected revenue forecast for the third quarter and reported that free cash flow had collapsed 90% from a year earlier, a casualty of surging capital expenditures. The stock dropped more than 7% in after-hours trading, adding to a slide that has left shares down 11% for the year as of Wednesday's close.

The spending numbers themselves keep climbing across the industry. Meta raised the low end of its 2026 capex guidance by $5 billion, putting the range at $130 billion to $145 billion. Alphabet hiked the top end of its own guidance to $205 billion last week and reported negative cash flow for the first time. Microsoft pegged its year's capex at roughly $175 billion. Amazon reports its results Thursday.

Against that spending, investors want to know what Meta actually gets in return. Zuckerberg's AI strategy has looked disjointed at times, and the company has fallen behind OpenAI, Anthropic and Google in the race for top-tier models and enterprise AI services.

"I think everyone wants clarity into what he wants to do in the compute business," Jefferies analyst Brent Thill told CNBC's "Closing Bell Overtime."

The trade-off Zuckerberg described

Rather than announce a product, Zuckerberg walked through the logic he's weighing. "In terms of running the business, obviously, a common trade-off that we need to make is around how much do you monetize something today versus develop future assets," he said. "I think that it's always a portfolio."

He framed a potential enterprise push as more than just renting out spare servers. Meta could also sell API access, productivity tools, and AI agents it's building internally, he said, calling it "a very, very large opportunity."

But he also pushed back on the idea of treating compute as a quick cash grab. Meta still needs plenty of capacity for its own ambitions, particularly as AI chief Alexandr Wang pushes out new models. Earlier this month, Meta released Muse Spark 1.1, which Wang described as the "strongest model for agentic and coding work yet," priced below comparable offerings from OpenAI and Anthropic.

"It would be foolish to basically just sell all of the compute and take a short-term profit," Zuckerberg said.

He also acknowledged that entering the enterprise market, an area where Meta has historically struggled, won't be simple. "That's going to be somewhat a new muscle that we build as a company," he said. "But I think it's a very important one that we build." Building that muscle likely means assembling a serious sales operation, though Zuckerberg didn't get into specifics. Meta has already brought in reinforcements: CNBC confirmed that Dave Brown, a longtime senior executive at Amazon Web Services, is joining the company.

A track record working against him

Zuckerberg's pitch to Wall Street runs into a familiar obstacle: his history of big, expensive bets that haven't paid off. The most glaring example is the metaverse, the 2021 pivot that aimed to remake Meta around virtual worlds. That effort is still burning cash. Reality Labs, the division building VR devices and wearables, lost $4.62 billion last quarter on just $431 million in revenue.

Still, the pressure to diversify is real. Digital advertising accounts for 98% of Meta's revenue, and AI represents the company's clearest shot at building something else at scale — assuming it can turn a massive infrastructure investment into an actual business rather than a sunk cost.

Zuckerberg, for his part, isn't hedging on the broader wager, even if the specifics of the compute strategy remain fuzzy. "I get that this is sort of a big bet across the industry," he said. "My personal bet is that the people who invest in this are going to be rewarded and feel very good over time."

What Meta hasn't done is put a name, a price, or a launch date on any of it. Until it does, investors are left parsing an earnings call for hints about a cloud business that, for now, exists mostly as an idea Zuckerberg is turning over in public.