Zuckerberg Wants to Sell Spare AI Computing Power, But Admits Meta Doesn't Know How Yet
Meta is sitting on enormous computing capacity and getting premium offers to lease it out. Zuckerberg says he's tempted, but selling too much now could kneecap the company's own AI plans.
Key points
- Meta raised its 2026 capital expenditure guidance to between $130 billion and $145 billion, a $5 billion increase at the low end.
- Mark Zuckerberg said on Meta's second-quarter 2025 earnings call that the company is receiving offers to lease spare computing power at prices above what Meta paid for it.
- AI startup Anthropic is in early talks to rent computing capacity from Meta, as first reported by CNBC Tech.
- Meta's free cash flow fell 90% from a year earlier as data centre costs soared.
- Advertising accounts for 98% of Meta's total revenue, which is why the company is searching for a second business.
Meta has quietly become one of the biggest buyers of AI computing power on the planet, spending between $130 billion and $145 billion this year on data centres and the specialised chips that run artificial intelligence. It has no business that sells any of that power to others.
Amazon Web Services, Microsoft Azure and Google Cloud all rent out computing capacity to businesses that don't want to build their own infrastructure. Meta, despite spending at similar scale, has no such product. Our report on Microsoft's record quarter noted that Azure revenue crossed $100 billion for the full year, a benchmark that underlines how far ahead the established cloud players already are from Meta's standing start.
Could Meta really sell computing power to other companies?
Zuckerberg says yes, but he's being careful. He confirmed on Wednesday's earnings call that Meta is fielding offers from outside companies willing to pay a premium above what Meta paid for its own capacity. Anthropic, the AI safety company behind the Claude family of chatbots, is reportedly in early talks to lease some of that power.
Still, Zuckerberg isn't ready to auction off spare capacity to the highest bidder.
"It would be foolish to basically just sell all of the compute and take a short-term profit," he told investors.
His reasoning is straightforward. Meta needs that processing power for its own AI models, including Muse Spark 1.1, a model its chief AI officer Alexandr Wang describes as Meta's strongest yet for coding and automated tasks. We've tracked Muse Spark across five stories since 13 July 2026, and the pace of releases suggests Meta is leaning hard on the model family internally before it considers licensing capacity out. Selling too much too soon could leave Meta unable to run its own systems at full speed.
What would a Meta cloud business actually look like?
Beyond raw capacity, Zuckerberg sketched a broader picture. He said a Meta enterprise offering could include API access (a way for outside software to connect directly to Meta's AI tools), productivity services, and AI agents (software that carries out multi-step tasks autonomously, like researching a topic or drafting documents).
"I think there's just a very, very large opportunity there," he said, without giving a launch date or specific product details.
He also acknowledged the obvious gap: Meta has never successfully sold to businesses. Building an enterprise sales operation is not something Meta has done before. Dave Brown, a former senior executive at Amazon Web Services, is joining Meta, a hire that suggests Zuckerberg is serious about filling that gap.
"That's going to be somewhat a new muscle that we build as a company," Zuckerberg said.
Should investors and users be worried?
Investors reacted badly. Meta shares fell more than 7% in after-hours trading after the company posted a weaker-than-expected revenue forecast and confirmed free cash flow dropped 90% year on year. As we reported the same evening, a wave of media appearances by Zuckerberg did nothing to slow the slide.
For ordinary users, the more immediate question is what Meta does with all this AI capacity. Consumer products built into WhatsApp and Instagram depend on it. If Meta leans too hard into renting capacity out, its own tools could slow down or improve less quickly.
Zuckerberg's track record adds legitimate caution. His last major bet beyond advertising was the metaverse, a virtual-reality digital world he pushed from 2021. Reality Labs, that division, lost $4.62 billion in the most recent quarter on just $431 million of revenue.
The AI computing bet is far larger than the metaverse was. My read: the Anthropic talks are real signal, not noise, but Meta is at least two years from a coherent enterprise cloud product, and the spending clock is running now. Whether hiring one AWS veteran and floating API deals counts as a strategy, or just buys time with Wall Street, is the question Zuckerberg has not yet answered.
Common questions
What does "computing capacity" mean in this context?
It means the physical servers and specialised AI chips inside Meta's data centres. When companies talk about leasing capacity, they mean letting outside organisations run their software on that hardware, charged by the hour or by volume of work processed.
Does this affect how Meta handles my personal data?
Not directly. The capacity being discussed is computing power for AI workloads, not user data storage. Meta's data policies would remain separate from any commercial cloud arrangement it strikes with other companies.
Why does 98% advertising revenue matter?
Almost every dollar Meta earns comes from one source: selling ad space on Facebook and Instagram. If ad spending drops, or if regulators restrict targeted advertising, Meta has almost nothing else to fall back on. A cloud computing business would give it a second revenue stream.
