Meta explores renting unused AI computing power to businesses
Plan is to monetise unused computing resources
Meta is developing plans to launch a cloud infrastructure business that would allow companies to rent AI computing power and access its AI models.
According to a report by Bloomberg, Meta is exploring ways to commercialise the vast computing infrastructure it has built for its own AI ambitions, potentially creating a new source of revenue from billions of dollars invested in datacentres and specialised AI chips.
One proposal under consideration would allow developers to access Meta-hosted AI models through an application programming interface (API), similar to services already offered by the hyperscalers.
Meta would operate the underlying infrastructure, including the datacentres and processors powering its own AI systems, and charge customers based on usage.
The company is also evaluating a second offering that would provide customers with direct access to raw computing capacity, enabling businesses to run their own AI workloads without relying on Meta's proprietary models.
Such a model would resemble services offered by neocloud specialists like Nebius or CrusoeAI or AI-natives like Coreweave and NScale.
The initiative forms part of an internal programme known as Meta Compute, which oversees the company's expanding AI infrastructure.
Bloomberg reported that the effort is being led by Meta's head of infrastructure, Santosh Janardhan, alongside Daniel Gross of the company's Superintelligence Labs and Meta president Dina Powell McCormick.
The plans remain under development.
AI infrastructure spending
Meta has increased spending on AI infrastructure over the past two years as chief executive Mark Zuckerberg pushes to make AI a central pillar of the company's future.
The investments include new datacentres, advanced graphics processors and large-scale computing agreements with infrastructure providers.
Those expenditures have raised questions among investors about how Meta intends to generate returns on its growing AI investments.
Entering the cloud computing market could provide one answer by allowing the company to monetise unused computing resources while demand for AI infrastructure continues to outstrip supply.
Renting AI computing capacity
The market for renting AI computing capacity has expanded rapidly as companies race to build and deploy increasingly powerful AI systems.
Cloud providers have invested heavily in specialised hardware and software platforms capable of supporting AI training and inference, creating businesses that generate tens of billions of dollars in quarterly revenue.
Building a competitive cloud platform, however, involves more than datacentres. Providers must also develop enterprise software, customer support operations and sales organisations capable of serving corporate clients at scale.
Zuckerberg has previously acknowledged that selling excess computing capacity remains a possibility if Meta builds more infrastructure than it ultimately requires.
Speaking to investors earlier this year, he said external companies had repeatedly approached Meta about purchasing computing capacity or accessing its AI capabilities through an API service.
While Meta currently expects to use most of its infrastructure internally, he said commercialising surplus capacity remained an option if the company found it had overbuilt.
Profiting from AI demand
The move would place Meta alongside a growing number of technology firms seeking to profit from the intense demand for AI infrastructure.
Earlier this year, Elon Musk's AI company xAI agreed to lease computing capacity from one of its large-scale datacentres to AI developer Anthropic after shifting its own training workloads to newer facilities.
Industry analysts expect demand for AI computing power to remain strong as companies continue investing heavily in increasingly sophisticated AI models, making access to large-scale computing infrastructure one of the fastest-growing segments of the technology industry.
Computing says:
There is a contradiction at the heart of these proposals. Very simply, if demand for AI compute is as strong as all the companies supplying it have been telling us it is for the last four years, and which has justified the billions of dollars of CapEx required to create that capacity, why has Meta got so much spare capacity for rent?
What lies behind these proposals is the fact that over the last decade, Meta has spent billions but delivered very little that people want. The metaverse? Glasses that creep everyone out? No thanks. At the same time, Meta has remorselessly pursued a strategy of maxmising ad revenue but making the user experience so bad that Facebook and Instagram are now borderline unusable by real people.
Meta has made a series of poor business decisions to the extent that it is now reliant on criminals for around 10% of its annual revenue. Meta will spin these proposals as evidence of insatiable AI demand. The (non-virtual) reality is that it’s Meta’s only hope of some sort of return on the money it’s spent chasing future returns from AI.