Meta enters cloud computing, will sell excess AI GPU capacity

Key Takeaways

• Meta will sell excess GPU capacity to outside customers, mirroring SpaceX's profitable compute rental model
• SpaceX earns $2.17 billion monthly renting GPUs to Anthropic and Google alone
• Meta's stock rose 10% on the news, despite questions about why it has spare capacity at all

Meta is building a cloud business to sell spare AI computing power to outside customers, Bloomberg reports. The company will rent out GPU capacity it originally purchased to train its own AI models. Meta's stock jumped roughly 10% on the announcement.

The strategy copies SpaceX almost exactly. Elon Musk's xAI division has been renting GPU capacity to competitors, pulling in $1.25 billion per month from Anthropic and $920 million per month from Google. That's $2.17 billion monthly from just two customers.

Why does Meta have spare compute?

This is the uncomfortable question buried in the announcement. Meta is one of Nvidia's largest GPU customers. The company laid off thousands of employees and committed to spending up to $145 billion on AI infrastructure this year alone. All that hardware was supposed to power better in-house models.

If Meta has capacity left over, the original thesis, that it needed massive compute to catch up in AI, looks weaker. The company's first model under Alexandr Wang, poached from Scale AI, launched in April. Meta called it "the first product of a ground-up overhaul of our AI efforts." That overhaul apparently didn't require all the GPUs they bought.

Reselling spare compute is smart asset management. It also signals that Meta's internal AI demand hasn't scaled as fast as its infrastructure spending.

The SpaceX playbook in detail

SpaceX built xAI to pursue artificial general intelligence. To do that, it amassed GPU clusters. When those clusters sat idle between training runs, SpaceX started renting them out. Anthropic signed on. Google signed on. The revenue stream now rivals what some cloud providers generate.

Meta's version may go further. Bloomberg reports the company could also offer access to AI models running on its infrastructure, not just raw compute. That would put Meta in direct competition with AWS, Azure, and Google Cloud on the AI serving layer.

What this means for AI teams

More GPU supply on the market benefits anyone training or running models. Nvidia's dominance has created capacity bottlenecks. When Meta and SpaceX rent out excess hardware, they pressure prices downward. Startups with limited budgets get more options.

The catch: renting from Meta ties your workloads to a company that competes directly with most AI businesses. Meta runs Llama. Meta runs AI features on Facebook, Instagram, and WhatsApp. Any team considering Meta cloud needs to weigh compute access against strategic exposure.

The $145 billion question

Meta committed $145 billion to AI infrastructure spending this year. That figure drew skepticism when announced. Critics asked whether any company could productively deploy that much hardware that fast.

A cloud business answers the question sideways. Meta doesn't need to deploy all that compute internally. It can monetize the excess. Whether that was always the plan or a pivot remains unclear. Either way, the financial pressure drops.

Market implications

The 10% stock jump suggests investors like the news. A cloud business creates recurring revenue from sunk infrastructure costs. It also diversifies Meta beyond advertising, the metric Wall Street has pushed for years.

For Nvidia, the move is neutral to positive. Meta keeps buying GPUs. Whether Meta uses them internally or rents them out, Nvidia gets paid either way. The hyperscalers, AWS and Azure in particular, face new competition for AI workloads. Neither needed that.

Frequently Asked Questions

When will Meta's cloud business launch?

Meta has not announced a specific launch date. The Bloomberg report confirms the company is building the business, but timeline details remain undisclosed.

How much does SpaceX charge for GPU rentals?

SpaceX's xAI earns $1.25 billion monthly from Anthropic and $920 million monthly from Google. Specific per-GPU-hour rates have not been publicly disclosed.

Will Meta's cloud compete with AWS and Google Cloud?

Yes, if Meta offers AI model access alongside raw compute as reported. This would position Meta directly against hyperscaler AI serving products.

Can startups access Meta's cloud compute?

Meta has not announced customer eligibility criteria. SpaceX's deals are with large enterprise customers; whether Meta targets smaller teams remains unknown.

Source: The Decoder / Matthias Bastian