Mark Zuckerberg has spent years building one of the most expensive AI computing networks on the planet. Now he wants to rent it out.
Bloomberg reported on July 1 that Meta is developing a new initiative called Meta Compute, designed to let the company lease its excess AI infrastructure to outside clients. The move would put Meta in direct competition with AWS, Microsoft Azure, and Google Cloud, the trio that currently dominates the cloud market. Meta’s stock climbed roughly 9-10% following the announcement.
What Meta Compute actually means
Zuckerberg had flagged this direction before. During Meta’s Q3 2025 earnings call, he hinted at the possibility, and he reinforced it at the May 2026 shareholder meeting, where he noted that outside companies had been approaching Meta asking to pay for access to its computing resources.
Beyond raw compute rentals, the initiative could also give external clients access to Meta’s AI models, essentially turning the company’s internal AI research into a billable product.
Reports from July 17 added another data point: Meta is reportedly in discussions with Anthropic for a computing lease that could be worth $10 billion over two years.
The capex math that keeps investors up at night
Meta raised its capital expenditure guidance for 2026 to $125-145 billion, more than double the $72 billion it spent in 2025.
In April 2026, the company expanded its commitment to CoreWeave by an additional $21 billion under a take-or-pay agreement that runs through 2032. A take-or-pay contract means Meta pays whether it uses the capacity or not.
The skeptic’s read on all of this is straightforward: Meta is a company that still depends on its own AI compute for its core products. Selling excess capacity assumes there is excess capacity, but if Meta’s own AI demands grow, that surplus could evaporate before it ever generates meaningful cloud revenue.
Why this matters beyond Meta’s stock price
The competitive implications for existing cloud providers are real. AWS, Azure, and Google Cloud have spent a decade building their enterprise customer relationships and developer ecosystems. Enterprise sales cycles are long, procurement processes are slow, and switching costs are high.
The Anthropic talks suggest a different path. Rather than competing for general-purpose cloud workloads, Meta could position itself as a destination specifically for AI-heavy compute, where its infrastructure and model access create a differentiated offering.
What to watch: whether the Anthropic deal closes, how quickly Meta reports cloud-segment revenue, and whether other major AI labs follow Anthropic’s lead in leasing capacity from Meta rather than the established hyperscalers.