Meta Plans Cloud Business to Sell AI Compute as Meta Platforms Stock Jumps 9%

The plan behind the Meta platforms stock rally

Meta is developing a cloud infrastructure business that would sell access to AI computing power and hosted models, Bloomberg reported Wednesday, citing people familiar with the matter. The plan would put Meta in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud, the three companies that currently control most of the US cloud market. Meta platforms stock jumped as much as 10% intraday and closed up nearly 9% on the report, one of the stock's biggest single-day moves in months.

According to Bloomberg and follow-up reporting from Reuters and CNBC, Meta is still deciding between two approaches: letting developers rent access to AI models hosted on its own infrastructure, including its Muse Spark model, and paying for the compute needed to run them, or selling raw computing capacity outright the way neocloud providers do. Meta declined to comment on the report, and Reuters said it could not independently verify the details. The plans are still in development and could change before anything ships.

Why Meta needs another revenue lever

The timing lines up with a capex problem Meta has been struggling to manage. In April, the company raised its full-year 2026 capital expenditure guidance to a range of $125 billion to $145 billion, up from $115 billion to $135 billion, citing higher memory and component prices along with additional data center costs. That figure is more than Meta spent in 2024 and 2025 combined. Multi-year infrastructure contract commitments jumped by $107 billion in the first quarter alone.

Meta has also been cutting costs elsewhere to make room for that spending. The company cut about 8,000 jobs in May, roughly 10% of its workforce, in a move CEO Mark Zuckerberg told employees was a direct consequence of the AI infrastructure budget. Selling excess compute and model access to outside customers gives Meta a way to turn part of that spending into revenue instead of pure cost, which helps explain why Wall Street reacted so positively to a plan that is not yet finalized. For more on the headcount side of this story, see AETW's ongoing tracker on AI-driven layoffs.

CoreWeave stock and the neocloud providers take the hit

Not every stock benefited. CoreWeave stock fell 10.8% and Nebius dropped 12.4% on the news, as investors weighed the risk that Meta, currently one of the largest customers for neocloud providers renting out GPU capacity, could pull back on outside spending once it has its own commercial cloud arm. Gil Luria, managing director at D.A. Davidson, compared the move to what pushed SpaceX toward selling its own excess compute, and said the bigger impact would land on neoclouds like CoreWeave and Nebius rather than on the established hyperscalers.

Meta would not be the first AI buyer to become an AI seller. SpaceX, through xAI, signed a deal in early May for Anthropic to buy out all of the compute capacity at SpaceX's Colossus 1 data center, and has since signed similar leases with Google and Reflection AI. Large AI labs turning into compute resellers is becoming its own pattern this year, driven by the same math: massive fixed infrastructure costs are easier to justify if excess capacity can be resold rather than left idle.

Where AWS Bedrock fits into Meta's playbook

The version of Meta's plan built around selling access to hosted models looks a lot like AWS Bedrock, Amazon's service that lets developers call AI models from multiple providers without managing the underlying infrastructure themselves. If Meta builds a similar layer on top of its own Meta Compute infrastructure, it would give outside developers a reason to build directly on Meta's stack instead of routing through AWS, Azure, or Google Cloud, the three platforms that still handle most enterprise cloud AI spending in the US.

Meta has been building toward this for months. The company set up a division called Meta Compute in January 2026 to grow and manage its AI infrastructure business, with a goal of developing tens of gigawatts of compute capacity this decade. Since then, it has broken ground on a data center in Oklahoma, filed permits for 12 additional buildings at its campus in El Paso, Texas, and signed an agreement to develop a site in Beloit, Wisconsin. At Meta's May shareholder meeting, Zuckerberg said entering cloud computing was definitely on the table, adding that outside companies were approaching Meta almost every week to buy access to its AI models or spare compute.

The open questions ahead

• Whether Meta sells hosted model access, raw compute, or both is still undecided, and the report says the plans could change before launch.
• Meta has not confirmed pricing, timeline, or which customers it might target first.
• Big Tech's combined 2026 capex, including Meta, Microsoft, Alphabet, and Amazon, is projected to reach roughly $725 billion, meaning this story extends well past Meta alone.
• Whether a Meta cloud business actually improves margins depends on utilization, something outside observers cannot verify until Meta reports real numbers.

Sources