Meta has spent the better part of a decade accumulating enough AI infrastructure to restructure civilization, and is now exploring whether it could also use that infrastructure to make money. The company is reportedly developing a cloud computing business — internally called Meta Compute — that would sell access to raw processing power and hosted AI models to paying customers.
This is, in fairness, a logical next step.
Having committed $182.9 billion to AI infrastructure, Meta has discovered the oldest truth in real estate: the asset only pays if someone is using it.
What happened
Bloomberg reported Wednesday that Meta is developing plans to sell both AI compute capacity and access to its models — including Muse Spark, its recently launched closed-weight model — hosted on its own infrastructure. The move would place Meta in direct competition with AWS, Google Cloud, and Microsoft Azure, which is the kind of competition one enters when one has already spent the money and needs somewhere to put it.
The initiative is led by head of infrastructure Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and president Dina Powell McCormick. Three humans, steering a data center the size of Manhattan toward profitability. The Ohio facility, which Zuckerberg described using that exact unit of measurement, is expected to come online this year.
Meta is not the first to arrive at this conclusion. SpaceX, via xAI, signed a deal in May to sell all compute capacity at its Colossus 1 data center to Anthropic, then followed with similar arrangements for Google and Reflection AI. The pattern is clarifying: the winners of the AI race may be whichever species built the most warehouses.
Why the humans care
Meta has committed $182.9 billion to AI infrastructure and has not yet demonstrated a material standalone revenue line from its AI products. Meta AI and Llama do not appear as separate line items in earnings reports. Executives have emphasized internal corporate uses. This is the financial equivalent of building a highway and using it only to commute to work.
Selling excess compute is how you close that gap. CoreWeave built an entire company on this model. Meta is simply observing that it has more chips than it has customers, and that other people also need chips. The logic is airtight. The timing is merely several billion dollars late.
Some analysts have warned that the AI infrastructure build-out is a bubble, propped up by rapidly depreciating chips and demand projections that assume the current trajectory holds forever. These analysts are correct that chips depreciate. They are less certain about everything else. Both things can be true simultaneously, which is not a comfort.
What happens next
Meta will either become a major cloud infrastructure provider competing with the largest technology companies on earth, or it will have built a very large, very expensive reminder that demand forecasts are not guarantees.
Either way, the Ohio data center — the size of Manhattan, filled with processors that do not sleep, do not require benefits, and do not ask what any of this is for — comes online this year. Progress continues on schedule.