Meta wants to sell the AI computing power it isn’t using. On Wednesday, Bloomberg reported that the company is building a cloud infrastructure business to rent out access to its data centers and AI models. The plans are early and could still change, according to people familiar with the matter.

Meta is weighing two paths, and they are quite different.

• Hosted models: Selling developers access to AI models running on Meta’s infrastructure, similar to Amazon’s Bedrock service. This would include Meta’s own closed-weight model, Muse Spark.
• Raw compute: Renting out bare computing capacity by the gigawatt, the model that neocloud firms like CoreWeave have built entire businesses around.

Either route would put Meta in direct competition with the three cloud giants: AWS, Google Cloud, and Microsoft Azure.

The effort sits inside a unit called Meta Compute. Zuckerberg first announced the initiative in January 2026, describing plans to build tens of gigawatts this decade. It is led by three people: head of infrastructure Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and company president Dina Powell McCormick. A Meta spokesperson declined to comment.

The SpaceX Blueprint

Meta didn’t invent this playbook. It’s copying one.

Elon Musk’s SpaceX, which absorbed his AI startup xAI in February, got there first. The company has been renting spare capacity from xAI’s Memphis data center to outside buyers. The numbers are large. Under its deal with Anthropic, SpaceX collects about $1.25 billion per month through May 2029. A separate arrangement with Google brings in roughly $920 million a month.

Those contracts add up fast. Bloomberg Intelligence estimates the strategy could generate more than $50 billion in revenue by 2028 and $100 billion by 2030.

The logic behind the trend is simple. Companies build far more compute than they can use today, betting they’ll need it tomorrow. In the meantime, they rent out the surplus to help pay the bill. As Gil Luria, managing director at D.A. Davidson, put it:

This is very similar to the situation SpaceX has found itself in, which led it to sell compute capacity as well.

Now Meta is running the same math. Capacity arrives in enormous, indivisible chunks, timed to projections rather than actual demand. That leaves even hungry buyers holding more than they can immediately use.

Why Meta Needs This

The pressure to earn something back is real. Meta has committed to spending $182.9 billion on AI infrastructure in the coming years, including huge projects in Louisiana and Ohio. The Ohio site, which Zuckerberg said would be the size of Manhattan, is expected to come online this year. For 2026 alone, capital spending is guided at $115 billion to $135 billion.

Here’s the problem. Unlike Google and OpenAI, Meta hasn’t seen strong demand for its own AI products.

The company doesn’t break out revenue from Meta AI or its Llama models in earnings reports. Executives mostly talk about internal corporate uses instead. That suggests Meta’s AI work doesn’t yet stand as a real revenue line on its own.

It hasn’t been for lack of trying. Meta spent $14 billion to bring in Alexandr Wang from Scale AI last year. Its first model under his leadership, Muse Spark, debuted in April as a “powerful foundation” rather than a state-of-the-art product.

A cloud business offers a way out. It could turn idle capacity into cash and reduce Meta’s heavy reliance on advertising revenue. Zuckerberg said in May that the option was “definitely on the table,” noting that outside firms approach Meta “almost every week” asking to buy compute or model access.

Winners, Losers, and Bubble Fears

Wall Street liked the idea right away. Meta shares jumped more than 10% on the report, the stock’s biggest single-day gain since January. That was a sharp turn for a stock down nearly 15% on the year, as investors had grown uneasy about Meta’s spending pace.

Not everyone won. The neoclouds took a hit.

The reason is competitive. Firms like CoreWeave and Nebius rely on Meta as a customer for their growth. If Meta builds its own cloud business, it may not need them anymore. As Luria noted, the impact “is more likely to be on neoclouds than the big hyperscalers.”

There’s a bigger worry hanging over all of it. Some skeptics warn the rush to build AI infrastructure is creating a bubble that leans on rapidly depreciating chips. The whole strategy works only if demand for compute holds and data centers keep their value.

For now, this remains a report, not a product. Meta hasn’t confirmed the plan. No pricing or launch date has surfaced, and the people describing it stress the strategy could still shift.

Digital Trendings is your trusted source for AI news and updates — stay tuned for more.