Meta is reportedly in talks to lease computing power to Anthropic, the maker of the Claude AI assistant, in a potential deal worth as much as $10 billion over two years. The New York Times broke the story Friday, citing three people with knowledge of the discussions. And Wall Street's reaction was modest but telling: Meta's stock, already down more than 2% in a broader tech selloff, pared some of its losses on the news.

That small price move captures something bigger. Investors are starting to see Meta not just as a social media company that sells ads, but as something closer to a computing utility.

What's actually being proposed

Anthropic, which is preparing for an IPO, reportedly proposed the arrangement in June. Under the terms being discussed, Anthropic would pay Meta monthly over the two-year period for access to Meta's computing infrastructure, the same kind of powerful hardware that runs AI models and trains new ones. Both companies could exit the deal early, according to the Times, and the terms remain subject to change. The discussions are described as early-stage and may not result in any agreement.

Meta has not previously been in the business of selling computing power to outside companies. That's the complication. Building a commercial cloud operation from scratch takes more than hardware. It requires pricing structures, contracts, technical support, and sales teams. Amazon, Google, and Microsoft have spent decades building those muscles. Meta has not.

But Meta CEO Mark Zuckerberg has been signaling this ambition publicly. At the company's shareholder meeting in May, he said entering cloud computing was "definitely on the table," noting that companies were approaching Meta "almost every week" to buy access to its AI models or spare computing power. Earlier this month, Bloomberg reported that Meta was already building a cloud business to sell excess capacity and host AI models for developers.

Why this matters beyond the deal itself

The computing infrastructure underlying AI is expensive to build and, once built, expensive to leave idle. The companies racing to deploy AI, including startups like Anthropic, face a real bottleneck: there is not enough computing capacity to go around, and what exists is controlled by a small number of cloud giants.

That scarcity has created an opening for newer entrants. Firms called neoclouds, companies like CoreWeave and Nebius that rent out specialized AI computing hardware, have grown fast precisely because demand outstrips what Amazon, Google, and Microsoft can supply. Anthropic itself struck a separate deal in May to use the full capacity of Elon Musk's SpaceX Colossus data center in Memphis.

Meta sitting on vast computing infrastructure and looking to monetize it fits the same logic. The company has spent tens of billions of dollars building data centers and acquiring the chips that power AI. Leasing that capacity to competitors turns a cost into a revenue line.

For ordinary people, the immediate consequence is indirect but real. The more competition there is in the market for AI computing, the more downward pressure there is on the cost of AI services over time. Right now, the market is tight and controlled by a few players. A Meta entry, even a partial one, adds supply. More supply, over time, tends to mean lower prices and more access.

The deeper pattern is about who controls the infrastructure layer of the AI economy. Cloud computing turned out to be one of the most profitable businesses in the history of technology. Whoever owns the servers effectively sets the terms for anyone building on top of them. Meta, having spent years and billions building infrastructure for its own purposes, is now asking whether it can turn that investment into leverage in a market it doesn't currently compete in.

The answer, as of Friday, is still unresolved.