Mark Zuckerberg has poured an astronomical $182.9 billion into AI infrastructure, but until now, this "iron fist" has served only the internal needs of the Meta empire. That is changing: the company is preparing for a head-on collision with cloud giants. According to Bloomberg, a project codenamed Meta Compute is brewing within the corporation—an initiative to sell computing power directly to enterprise clients. The project is steered by heavy hitters: Santosh Janardhan (Head of Infrastructure), Daniel Gross (Head of Meta Superintelligence Labs), and Dina Powell McCormick, the company’s president. This is no longer just an experiment, but an attempt to transform a bottomless debt pit into a money-printing machine.
The Shift to Compute-as-a-Service
We are witnessing the rapid commoditization of data centers. Meta isn't the first to rent out "excess" capacity—Elon Musk has already set the trend. As Rebecca Bellan of TechCrunch notes, SpaceX effectively turned its internal resources into a lucrative business, striking a deal in May with Anthropic to buy out the entire capacity of the Colossus 1 data center. Zuckerberg is mirroring this maneuver. Unlike Google, Meta has struggled to create mass demand for its specific AI services, making the sale of "raw" hardware the fastest way to generate liquid cash here and now. Before the infrastructure begins to face moral obsolescence, it must be converted into a service.
"Meta's AI projects have yet to bring in tangible direct revenue, remaining merely a line item on the expense report."
Financial Necessity and Scale
The lack of profit makes Meta Compute a financial necessity. While top management previously touted the benefits of neural networks for ad targeting, Zuckerberg admitted in May that launching a cloud business is "definitely being discussed." The strategy bears a suspicious resemblance to the CoreWeave model: a focus on hardware rather than software. This will allow the company to monetize massive construction projects in Louisiana and Ohio. The Ohio facility, which Zuckerberg immodestly compared to the scale of Manhattan, is slated to go online this year. For investors, this is a clear signal: the company is trying to justify its bloated capital expenditures.
Market Pressure and the Bubble Risk
Entering the public cloud market forces Meta to solve problems it was previously shielded from—such as ensuring data isolation (multi-tenancy) for third-party businesses. While the company plans to follow in AWS's footsteps by offering hosting for various models, including its own proprietary Muse Spark, its main trump card remains the sheer volume of available chips. However, skeptics cited by TechCrunch warn that the market is in a bubble phase built on rapidly depreciating hardware. A reasonable question arises: will end-users be able to generate enough profit to pay off the trillion-dollar bets of these giants?
Zuckerberg is effectively going all-in on physical assets. In this arms race, the winner won't be the one with the smartest model, but the one who owns the "real estate"—the data centers. While Meta’s financial reports are silent on Llama revenues, direct capacity rentals turn a cost center into a potential profit center. This is excellent insurance in case the hype around specialized models cools down: the demand for compute will remain regardless. By the end of the first quarter, Meta had already sunk $182.9 billion into this foundation; now the only question is whether it can undercut Tier-1 providers by leveraging its already-sunk R&D costs.