Meta Plans Excess Compute Sales

Meta's possible move into cloud computing would turn part of its AI infrastructure buildout into a commercial service, but the revenue line would come with lower-margin economics than the company's advertising business.

Jim Cramer confirmed that Meta will sell excess computing power to outside customers.

Mark Zuckerberg had said at Meta's annual shareholder meeting in May that a cloud computing business was definitely on the table.

Zuckerberg also said on an earnings call seven months earlier that companies were regularly asking whether Meta had compute capacity they could buy at a premium to Meta's own cost.

The plan would put Meta closer to AI-specific cloud providers than to a full AWS-style hyperscaler.

Mark Mahaney of Evercore said Meta is unlikely to challenge Amazon Web Services, Microsoft Azure and Google Cloud directly, and instead may follow neocloud providers such as CoreWeave and Nebius.

Meta's Capex Increase Draws Investor Attention

Meta shares jumped 9% on Wednesday, their sharpest rally in more than five months, after investors reacted to the cloud-computing direction.

The move followed a difficult stretch in which the stock had closed out a fourth straight quarterly drop and lost almost a quarter of its value.

The capital-spending backdrop is large.

Meta's April earnings materials listed a $10 billion increase to the top of the 2026 capital expenditures range, bringing that upper target to $145 billion.

Some of that spending is being funded through debt, with the company raising $25 billion in a bond sale as it reported first-quarter earnings.

Karan Ramchandani, managing director at Post Oak Group, said turning compute into a revenue stream had been part of Meta's road map and described selling compute to other business customers as a logical market move.

Paul Meeks, head of technology research at Freedom Capital Markets, said the push responds to complaints that Meta may be overspending and to doubts over whether the company will earn a matching return on capex.

AI Compute Revenue Is Not Ad-Margin Revenue

Margins shape the economics.

Meta still gets 98% of its revenue from digital ads, and its gross margin was listed at 82%, with an operating margin of 41% in the latest quarter.

Google offers a lower-margin comparison.

Google's services business, which mostly comes from ads, recorded a 42% operating margin in the first quarter, while its cloud business posted an 18% margin.

The cloud margin took years to mature.

Google's disclosed cloud timeline says the infrastructure business launched in 2008, became generally available in 2011, began financial disclosure in 2020 and did not record a cloud profit until the first quarter of 2023.

Meta has also tested other monetisation routes.

Meta shares rose almost 4% in May when the company announced paid subscription plans for Instagram, Facebook and WhatsApp, along with two subscription services for its Meta AI app and website.

Those subscription efforts sit beside a larger infrastructure decision.

Selling cloud access would require enterprise sales and support work, while paid consumer or creator services use Meta's existing platforms.

Meeks said anything Meta enters outside online ads would dilute the business and lower margins from the company's strongest years.

He also said he would rather see Meta monetise AI through products and services with higher margins than enter what he described as a brutal data-centre buildout battle.

Neocloud Comparisons Depend On Customer Traction

Meta's possible cloud move sits beside companies that monetise AI-specific capacity rather than broad enterprise cloud services.

CoreWeave and Nebius both suffered double-digit share drops after the Meta report.

Mahaney said SpaceX may have partly influenced Meta's thinking.

SpaceX, which owns xAI, has capacity agreements with Google and Anthropic cited at more than $2 billion in combined monthly revenue, plus a separate deal with Reflection AI.

Brian Schechter, a partner at Primary Venture Partners, said Meta and SpaceX are similar because both spent billions of dollars training large AI models on their own infrastructure.

He said monetising compute after a missed training run shows that compute can function more like a commodity.

Customer proof remains unresolved.

Meta declined to comment, and no named cloud customers, signed contracts, pricing, service-level terms or launch date for the planned compute offering have been disclosed.