Meta Plans Cloud Business to Sell Excess AI Compute, Shares Up 10% Today

Meta's shares are up roughly 10% in Wednesday trading after Bloomberg reported the company is developing a cloud infrastructure business — internally dubbed Meta Compute — that would sell access to AI computing power and potentially AI models to outside customers.

The initiative is led by head of infrastructure Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and [Note: the article's sourced claim that "company president Dina Powell McCormick" is involved cannot be independently verified; Dina Powell McCormick is known as a Goldman Sachs executive, not a Meta president], according to Bloomberg.

Mark Zuckerberg telegraphed this move publicly. At Meta's Q3 2025 earnings call and again at the May 2026 annual shareholder meeting, he told investors a cloud business was "definitely on the table" if Meta found itself having overbuilt AI infrastructure.

The Numbers Behind the Push

As of the end of Q1 2026, Meta had committed to spending $182.9 billion on AI infrastructure over coming years, according to TechCrunch. CNBC reported the company told investors in April it plans to spend as much as $145 billion in capital expenditures in 2026 alone — primarily on data centers and the graphics processing units needed to train and run AI workloads.

Two massive data center projects are currently underway: one in Louisiana and one in Ohio. Zuckerberg has described the Ohio facility as the size of Manhattan, with an expected online date later this year.

That is an enormous infrastructure bet for a company that does NOT break out standalone revenue from Meta AI or its open-weight Llama model family in earnings reports. Executives have largely pointed to internal corporate productivity gains when discussing AI returns, not a revenue line that would justify the scale of spending on its own.

Competing With Giants, and Hurting Smaller Players

If Meta Compute launches, it enters one of the most competitive markets in tech: cloud infrastructure, currently dominated by Amazon Web Services, Google Cloud, and Microsoft Azure. CoreWeave, the GPU-focused neocloud that went public earlier this year, also competes in this space.

The market reacted immediately. Shares of CoreWeave and Nebius Group each fell roughly 12% Wednesday following the Meta announcement, according to CNBC. Investors clearly read Meta's entry as a threat to the smaller players who have been capitalizing on AI compute scarcity.

xAI Did This First

Meta is following a path xAI already walked. In early May 2026, xAI — Elon Musk's AI company — signed a deal with Anthropic to sell out all compute capacity at its Colossus 1 data center. Anthropic agreed to pay $1.25 billion per month for that capacity, according to CNBC. Google signed a separate lease for $920 million per month. xAI has since added Reflection AI to its customer list.

When the companies building the most AI infrastructure start selling unused capacity rather than consuming it internally, it raises a legitimate question about whether AI application demand has kept pace with infrastructure supply.

The Skeptic's Case

Not everyone sees this as a win. Some analysts have warned that the frenzy to build AI data centers is inflating a bubble built on rapidly depreciating hardware, as TechCrunch noted. The concern is straightforward: AI chips lose value quickly, data center construction is enormously capital-intensive, and end-user AI revenue — subscriptions, API fees, enterprise contracts — has not yet scaled to justify trillion-dollar infrastructure bets across the industry.

Meta is spending at a scale that dwarfs most sovereign nations' technology budgets, and it has not demonstrated a consumer AI product with breakout demand. [Note: The original article's claim that Meta's "newest model" is called "Muse Spark" and was "launched in April" under Alexandr Wang's leadership cannot be verified — Meta's known public model family is the Llama series, and "Muse Spark" does not correspond to any confirmed Meta model release.] Alexandr Wang — whom Meta paid roughly $14 billion to bring over from Scale AI last year — leads Meta's superintelligence efforts, but the specific model claims above require independent verification.

Selling compute to others, then, is not just a revenue opportunity. It may be a hedge against the possibility that Meta's own AI products don't generate enough demand to fill the infrastructure it has already committed to building.

What This Actually Means

If xAI and Meta — two organizations that poured resources into building AI capabilities — are now selling capacity rather than using it all internally, the scarce resource in AI may not be model quality. It may be physical infrastructure: power, land, fiber, and chips.

That repositions Meta less as a software-and-services AI company and more as a picks-and-shovels infrastructure play, closer to CoreWeave's model than to OpenAI's.

The unresolved question is whether Meta Compute can sign customers at rates that justify the spend — and whether the Ohio data center, expected online later in 2026, arrives into a market that still has more demand than supply, or one that has already started to tip the other direction.