Meta is expanding its Louisiana data centre to 5 gigawatts, roughly double its planned capacity, making it the company's largest site. Mark Zuckerberg has confirmed he is weighing whether to rent that capacity out. "We're looking at that," he told Bloomberg. Put alongside what Meta has been building in silicon, the question writes itself. Is this the beginning of an AWS for the AI era?
The pieces are all on the board
The comparison is tempting because Meta now owns the full stack. It has its own chips, the Meta Training and Inference Accelerator line, or MTIA, with four designs unveiled in March and a new one shipping about every six months. It has one of the largest Nvidia fleets in the world. It has Llama, the most downloaded open-weight model family. It has 33 data centres built or in development, and capital spending guided at $125bn to $145bn for this year.
Bloomberg reported on 1 July that Meta is planning a cloud business internally called Meta Compute, selling idle computing power and model access to outside customers. Two paths are described. Host Meta's own models for developers, which would look like Amazon's Bedrock. Or lease bare capacity, which would look like CoreWeave. Zuckerberg told investors in May that companies approach Meta most weeks asking to buy compute, some offering to pay above what it costs Meta to procure.
Why the AWS analogy flatters
Here the comparison starts to strain. The founding story that AWS was simply Amazon's spare capacity resold is mostly myth. AWS was built as a product, with pricing, support and a sales organisation behind it. That is the part Meta has never done. Enterprise cloud runs on service-level agreements, procurement cycles and sales engineers. Meta has no history of any of it.
The silicon shows the gap. Google and Amazon put their custom chips inside their clouds so customers can rent them. MTIA has no external availability, no partner programme and no migration path for anyone outside Meta's own software stack. Google's TPU took years to become something a stranger could use. Meta would be starting that clock now.
The question underneath
If Meta does go ahead, the consequences reach past Meta. The July report alone knocked more than 10% off the Philadelphia Semiconductor Index over two sessions. The reason is that a new seller of compute attacks the belief the whole AI trade rests on, that compute is permanently scarce. Idle capacity for hire is the opposite of scarcity.
That is also the tell. Meta is not renting capacity because demand for it is overwhelming. It is considering renting because it built more than it can use, and because its own AI products have moved slower than the buildout. Selling the surplus turns a fixed cost into revenue and hedges a spending programme investors already find heavy.
So the honest answer is not yet, and perhaps not in the way the analogy suggests. AWS created a market. Meta would be entering one where the largest buyers are already committed to Amazon, Microsoft and Google. Zuckerberg has the hardware, the models and the power. What he does not have is a customer list, and the next earnings call is where he would have to name one.
