Meta wants to rent out AI capacity and take on AWS and Google Cloud

After spending the past two years stockpiling AI computing capacity on a massive scale, Meta is considering renting out its surplus to third-party companies. If the plan goes ahead, it will put Meta in direct competition with the cloud giants AWS, Google Cloud and Microsoft Azure for the first time, and Wall Street has already reacted positively to the idea.

Where all the spare capacity came from

Starting around 2024, Meta aggressively expanded its AI infrastructure. The company bought Nvidia H100 GPUs and in parallel developed its own MTIA AI chips. These are needed to train the Llama language models and to optimize the advertising algorithms that generate most of its revenue. Data centers were built with headroom, ahead of current demand: the infrastructure was meant to grow alongside the company's ambitions in generative AI.

As a result, part of the capacity it bought sits underused. According to sources, Meta is studying a cloud-provider business model in which corporate customers could rent compute to train their own AI models or run inference. That would turn hardware that is already paid for into a source of revenue without requiring additional investment.

Key facts:

• Meta has been building up AI infrastructure for at least two years
• Potential direct competitors are AWS, Google Cloud and Microsoft Azure
• Wall Street reacted positively to the idea
• The core of the plan is monetizing idle capacity, not building a cloud platform from scratch

Why investors like it

Meta's capital spending on AI infrastructure has long drawn mixed feelings from analysts. Every quarter the company reports multibillion-dollar outlays on GPUs and data center construction. The market accepts that AI is a strategic priority, but a concrete return on that investment has long remained distant and unclear. Any scheme that turns heavy assets from ballast into a source of direct revenue changes that narrative significantly.

Amazon, Google and Microsoft earn hundreds of billions of dollars a year from cloud services, and Meta is unlikely to break into that market quickly as a full-fledged provider: that takes a multi-year ecosystem of enterprise products and established relationships with the IT departments of large companies. In AI compute specifically, however, meaning GPU clusters for training and inference of large models, demand chronically exceeds supply, and companies around the world wait months for access to capacity. This is where Meta has a real competitive advantage: its own global network infrastructure, custom AI chips and a scale comparable to the hyperscalers.

The model itself is not new: the specialized providers CoreWeave and Lambda Labs have already made GPU rental their core business. The key difference for Meta is that it has no traditional enterprise cloud business behind it, only huge idle capacity that needs to be monetized.

What this means for the market

If the project goes ahead, Meta will go from the largest consumer of AI infrastructure to one of its sellers. Companies without hyperscaler resources would gain another major GPU compute supplier, potentially on different terms than AWS or Google Cloud. More competition in this segment puts pressure on prices over the long term and widens access to AI capacity for a broader range of players.

For Meta itself, it is a chance to change the narrative: to turn the story of a "company that spends billions on AI" into the story of a "company that makes money from AI."

*Meta is designated an extremist organization and banned in Russia.