Meta Announces Cloud Business to Sell Excess AI Compute
Meta has announced plans for a new cloud infrastructure business, designed to offer outside customers access to AI models and computing capacity.
These plans would see Meta compete with the likes of other tech company services like Amazon Web Services, Microsoft Azure and Google Cloud.
An internal group, named Meta Compute, will be instrumental in these efforts. The segment was designed to oversee the buildout and operations of Meta’s AI infrastructure.
Meta Compute is run by several company executives, including Santosh Janardhan, Meta’s Head of Infrastructure; Daniel Goss, an executive within Meta’s Superintelligence Labs AI Unit; and Meta President Dina Powell McCormick.
One of the options under consideration for the computing segment is a service model that would allow outside developers to pay to run queries against AI models – including Meta’s own Muse Spark – on infrastructure that the company owns and operates.
Additionally the company would create a separate avenue to rent out raw GPU capacity directly to Meta customers.
Selling computing access externally
Speaking to shareholders in May, Meta CEO Mark Zuckerberg said the idea of selling computing access was something the company had been considering, adding that the idea was “definitely on the table”.
He added: “Almost every week there are different companies that come to us from the outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we've bought it at.”
He went on to say that no outside deals had been made as internal demand had so far taken up the available capacity. However, Mark added that if overbuilding was to occur, it would move to sell the extra capacity externally.
Investing into AI superintelligence
Superintelligence has become a major priority for Meta, with the company investing hundreds of billions of dollars into AI infrastructure and reaching large capacity agreements with CoreWeave, Google and Oracle.
While some of Meta’s investors question how operations like this will translate into revenue, the potential cloud business could prove a profitable venture.
The compute supply constraints that have been a key point of discussion with Meta’s AI strategy first came into focus earlier this year when Google restricted Meta’s access to its Gemini AI. Google said that it was unable to meet Meta’s demand for AI compute.
This delayed some of Meta’s internal AI efforts and resulted in Meta asking its own employees to reduce their AI token consumption.
Muse Spark, Meta’s newer proprietary model, has since taken on a large share of the work initially handled by Gemini, as Meta starts to develop and grow its own in-house AI capabilities.
The model was first unveiled in April, but is yet to be released to developers. According to a report by The Wall Street Journal, a confirmed release date for the AI model is yet to be announced.
Meta is projected to spend as much as US$145bn on AI infrastructure this year, a significant portion of the tech industry’s average spend of US$700bn on the technology.
