Meta is considering selling access to its growing AI computing empire

Signals Inbox·July 10, 2026·AI Infrastructure

Meta built its enormous AI infrastructure for its own models, ads and products. Now Zuckerberg is considering renting some of that computing power to outsiders, potentially turning one of Meta’s biggest expenses into a new cloud business.

Q1What did Zuckerberg actually say?

At Meta’s official shareholder meeting, Zuckerberg said a cloud computing business was definitely on the table. Companies regularly approach Meta about buying access to its models or computing power. Meta has not officially launched the service yet.

Q2What would Meta actually rent?

Basically, access to the expensive machines behind modern AI. That could include raw GPU computing power, hosted access to Meta’s AI models, or both. A startup could pay Meta to train or run AI instead of buying its own chips and building a data center.

Q3Why would Meta rent compute when it needs so much itself?

That is the tension. Zuckerberg says Meta can currently use all the compute it gets. But AI capacity is hard to plan perfectly. If Meta occasionally has more capacity than it needs, or receives an unusually high offer, renting it could make more financial sense than keeping every machine reserved internally.

Q4Why now?

Meta expects to spend roughly $125 billion to $145 billion in 2026, with AI infrastructure driving much of it. At that scale, investors want to know how the spending becomes revenue. Renting compute would give Meta a second answer beyond better ads, smarter assistants and stronger consumer products.

Q5Who would feel the most pressure?

Probably specialist AI cloud companies such as CoreWeave and Nebius before AWS itself. Their pitch is simple: we can get you scarce GPUs quickly. Meta owns a huge fleet, has deep pockets and could sometimes rent capacity that was originally purchased for internal use.