Meta Cloud Computing Business Turns $145 Billion Into Rent

Meta Platforms told investors in April to expect capital spending of $125 billion to $145 billion in 2026, nearly all of it for AI data centers and computing hardware. On July 1, the market learned that the bill may soon start paying for itself. Bloomberg News reported that the company is drafting plans for a Meta cloud computing business, known internally as Meta Compute, that would sell outside customers access to its AI computing power and its AI models. Meta shares climbed about 9% on the day by CNBC's count, while Reuters-syndicated coverage put the intraday gain at more than 10%.

A Meta cloud computing business would end the company's standing as the only major U.S. hyperscaler without a commercial cloud arm, and it would open a direct front against Amazon Web Services, Microsoft Azure and Google Cloud. Reuters confirmed Bloomberg's report the same day. The plans remain in development, with no launch date, pricing or customer pipeline disclosed, and Bloomberg cautioned that the strategy could still change.

The stakes are considerable even by hyperscaler standards. AWS generated roughly $115 billion in revenue in 2025 and Google Cloud about $44 billion, while Big Tech's combined spending on AI in 2026 is estimated to exceed $700 billion. Against that backdrop, a fourth full-scale entrant with Meta's balance sheet would redraw the competitive map of cloud infrastructure.

The Last Holdout Reverses Course

The executive lineup signals how seriously Meta is treating the effort. Bloomberg reported that the project is led by Santosh Janardhan, the company's head of infrastructure; Daniel Gross of Meta Superintelligence Labs; and Dina Powell McCormick, Meta's president. Assigning three senior leaders to a business that does not yet have a launch date suggests the company views Meta Compute as a strategic priority rather than an experiment.

The decision has been telegraphed for weeks. At Meta's shareholder meeting in May, Chief Executive Mark Zuckerberg said companies were approaching Meta "almost every week" seeking access to its AI models or spare computing power, and that entering cloud computing was "definitely on the table." He added that Meta had not sold access to that point because it expected to use the capacity internally.

That caveat matters for the analysis. A cloud business built on genuinely surplus capacity is close to found revenue; one that diverts compute from Meta's own model training and advertising systems carries a real opportunity cost. The company has not said which it would be, and the answer will determine how much of the reported plan is pure monetization and how much is a bet that external demand is worth more than internal use.

Two Routes to Market, Neither Priced

According to Bloomberg, Meta is evaluating two business models. The first is hosted access to AI models through APIs, a managed service comparable to Amazon's Bedrock. The second is rental of raw GPU computing capacity as infrastructure-as-a-service, the commodity end of the cloud market where price per chip-hour dominates.

The two paths carry different economics and different competitors:

• Hosted model APIs would give Meta a direct revenue channel for its AI models, including the Muse Spark model unveiled in April 2026, and would compete on model quality as much as on price.
• Raw GPU rental would pit Meta against the established hyperscalers and against specialist GPU clouds, in a segment where scale and utilization decide margins.

There is precedent for AI labs renting out spare capacity. Coverage of the Meta plan cited xAI, which has rented out data-center capacity in Memphis. The difference is one of scale: Meta's 2026 capex guidance alone exceeds the roughly $115 billion in revenue AWS produced last year.

Suppliers Turned Rivals in a Single Session

The sharpest market verdict on July 1 came not from Meta's own stock but from its suppliers. CoreWeave fell 10.8% and Nebius dropped 12.4%, as investors concluded that a customer that has anchored the specialist GPU-rental sector could become its most dangerous competitor.

The reversal is especially pointed for CoreWeave. Meta expanded its infrastructure agreement with the company to $21 billion in April 2026, covering capacity through December 2032. That contract now cuts both ways. It locks in revenue for CoreWeave for six more years, and it hands Meta a supply cushion it could effectively resell into the very market CoreWeave serves.

The selloff exposed a structural weakness in the so-called neocloud model. Firms such as CoreWeave and Nebius exist largely to absorb Big Tech's overflow demand for GPUs. If the hyperscalers that generate that overflow begin selling their own spare capacity, the neoclouds face compression from both directions: their largest customers need less third-party capacity, and their prospective customers gain a new, deep-pocketed supplier. The scale of the one-day declines suggests investors had priced little of that risk.

Wall Street Accepts a Margin Trade

Analyst reaction through July 2 and 3 was broadly positive despite uncomfortable arithmetic. CNBC reported on July 2 that Wall Street was enthusiastic about the growth opportunity even though a capital-intensive cloud business would carry structurally lower margins than Meta's advertising operation. Cloud infrastructure demands continuous hardware refresh cycles, data-center construction and power procurement, costs the advertising business does not bear at anything like the same intensity.

Commentary in Forbes and The Motley Fool on July 2 and 3 framed the plan as the clearest answer yet to the question that has trailed Meta all year: whether the company's record AI outlay can pay for itself. Until now, the $125 billion to $145 billion budget has been justified by internal returns, meaning better ad targeting, new AI products and progress toward more capable models. A cloud business would convert part of that budget into a product line with its own revenue, shifting record capex from cost center to potential profit engine.

Several critical details remain undisclosed, and each will shape whether the enthusiasm holds:

• No launch date, pricing structure or customer pipeline has been announced.
• Meta has not said how much capacity it considers surplus, or how it would arbitrate between internal and external demand for the same chips.
• Bloomberg reported that the strategy could still change before any product reaches the market.

For now, the reported plan moves the Meta cloud computing business debate from whether the company should enter the market to how, and how fast, it will.

The competitive consequences arrive on two clocks. For AWS, Azure and Google Cloud, a Meta entry is a medium-term threat contingent on execution in a business Meta has never run. For CoreWeave, Nebius and the rest of the GPU-rental specialists, the threat repriced in a single afternoon. Whatever Meta Compute ultimately becomes, July 1 established that the market now treats Meta's $145 billion buildout as a product in waiting rather than a cost to be endured.