The strategy test: can Meta turn the AI infrastructure it built for its own products into a third party cloud business, the way SpaceX has begun renting its Memphis data center to outside AI labs?
Meta shares climbed as much as 12% to about $628 on Nasdaq on Wednesday after a Wall Street Journal report that the company is preparing an external AI cloud offering under the internal program name "Meta Compute." The price level and intraday move come from Livemint's wire copy of the report, with Reuters and The Verge also reporting on the same WSJ scoop.
The market was not just buying an AI story. It was buying the possibility that Meta can do what SpaceX has been doing with its Memphis data center, internally known as Colossus 1, and rent out the AI infrastructure it has been building for itself to outside customers. That would convert a cost center into a product.
The shape of the offering, as the WSJ report describes it, is one of two options. Meta could host access to its own AI models, the equivalent of Amazon Web Services' Bedrock platform for running third-party AI workloads on rented infrastructure. Or it could rent out raw compute and AI chips directly from its data centers, the way a commodity cloud provider does. Either path would put Meta against Amazon Web Services and Microsoft Azure, the two cloud incumbents that already rent AI compute to enterprise customers at scale, with Google Cloud a step behind.
The Verge's read is that the bet is structural: Meta would be leveraging the same AI infrastructure it has been buying from Nvidia, AMD, and Broadcom for its own products and trying to sell the excess to someone else. TechCrunch's framing makes the SpaceX comparison explicit. SpaceX built out the Colossus 1 facility in Memphis to train its own models, then began renting capacity to outside AI labs, including Anthropic. That is the operating model Meta is reportedly testing.
The capex backdrop is the reason investors took the report seriously. Big Tech's AI infrastructure capex is on track for a record year, anchored on guidance from the largest US cloud and tech companies, and most of that spend only pays back if someone other than the company that built the data centers is paying the bills. The market read on Wednesday was that Meta has both the chips and the data center footprint to make that bet, and the strategic willingness to do it.
Zuckerberg had already telegraphed the move. In a May 2026 CNBC interview, he said launching a cloud computing business was "definitely on the table" and pointed to monetizing excess AI compute and paid AI services as the natural next step. The WSJ report suggests that step is closer than investors had assumed.
What is still unknown is whether Meta Compute would be a real revenue line or a marketing layer over Meta's existing AI offerings. The source register is "exploring," "considering," and "one option under consideration." There are no named customers, no launch date, and no revenue impact. The 12% move is a market bet, not a product announcement. The test for Meta Compute is the same one SpaceX passed with the Colossus 1 rentals and that most internal-to-external infrastructure conversions have historically failed: it has to clear a price, a service-level agreement, and a go-to-market that third parties will actually pay for, not just a slide deck that convinces Wall Street for an afternoon.