Meta Wants to Sell You Compute. The Irony? It Couldn't Buy Enough of Its Own.

Meta is reportedly building a new business called Meta Compute, and the pitch is bold: rent out AI model access and raw GPU capacity to outside developers, going head to head with AWS, Azure, and Google Cloud.

On paper, this looks like classic Big Tech expansion. In practice, it's a fascinating reversal of Meta's own recent history.

Not long ago, Meta was on the other side of this exact transaction, and it didn't go smoothly. Google reportedly couldn't meet Meta's demand for Gemini compute, a shortage serious enough that Meta had to ask its own employees to cut back on AI token usage internally. That's not a hypothetical capacity problem. That's a company building superintelligence-scale ambitions running into a very physical wall: there simply wasn't enough silicon to go around.

Now Meta wants to be the supplier instead of the customer.

The strategy has two tracks. First, external developers could pay to query Meta's proprietary model, Muse Spark, the same system that's reportedly been quietly absorbing workloads that Gemini used to handle. Second, and more interesting, Meta could rent out raw GPU capacity directly, essentially becoming an infrastructure landlord the same way AWS built its empire.

CEO Mark Zuckerberg told shareholders in May that outside companies "almost every week" ask Meta to either stand up an API service or sell them compute at a markup. His answer so far has been no, not because the demand isn't there, but because internal appetite has swallowed everything Meta has built. His stated logic: if Meta ends up overbuilding capacity, selling the surplus becomes the obvious move.

And Meta is not underbuilding. The company is projected to spend up to $145 billion on AI infrastructure this year alone, a meaningful slice of the roughly $700 billion the entire tech industry is expected to pour into AI this year.

Here's the tension nobody's fully resolved yet: investors are already asking how a compute-reselling business actually turns a profit for a company whose core business has always been advertising, not infrastructure rental. AWS took Amazon over a decade to become a dominant, high-margin business. Meta is entering a market where Amazon, Microsoft, and Google already have years of enterprise trust, global data center networks, and established billing relationships with the exact developers Meta hopes to win over.

So is Meta Compute a genuine third act for the company, or a defensive hedge in case its superintelligence bet needs new revenue streams to justify the spending? The honest answer is probably both.

Would you trust your production workloads to a cloud provider whose day job has always been social media?