Google's AI demand is so high that it's renting supercomputers from SpaceX

Google has signed a deal to rent AI computing infrastructure from SpaceX. The agreement underlines how demand for AI capacity is outpacing new data centre build-outs.

In Short

• Google is renting AI infrastructure from SpaceX to meet demand
• AI computing demand is growing faster than new capacity
• SpaceX is turning excess AI infrastructure into a major business

Demand for AI computing power is rising so rapidly that even companies with massive budgets are scrambling to find additional capacity wherever they can. Instead of building everything themselves, some AI firms are now renting infrastructure from other companies to keep up with demand. The latest example is Google.

SpaceX announced in a regulatory filing that Google has signed a massive agreement to rent AI computing infrastructure from the company. Under the deal, Google will pay SpaceX $920 million per month from October 2026 through June 2029 for access to approximately 1,10,000 NVIDIA GPUs, CPUs, memory and other related components.

To put that into perspective, Google is not buying the hardware. It is effectively renting access to a giant pool of AI computing power because demand for its AI products is growing faster than infrastructure can be deployed.

Why does Google need someone else's computers?

At first glance, the deal may seem surprising. Google is one of the world's largest technology companies and has already committed more than $180 billion in capital expenditures this year. The company has also said that spending will increase significantly in 2027 and recently announced an $80 billion equity sale to help fund its expansion plans.

Yet even that may not be enough.

According to Google, demand for its AI services has exceeded expectations. In a statement to TechCrunch, the company said the agreement would help meet surging customer demand for its AI offerings, including its agent platform and Gemini Enterprise. In other words, Google needs computing power now, while building new data centers can take years.

The new gold rush isn't AI models, it's AI infrastructure

The deal highlights a growing reality in the AI industry: computing power has become one of the most valuable resources in technology. Training and running advanced AI models requires enormous numbers of specialised chips, particularly NVIDIA GPUs. As AI adoption accelerates, demand for these chips and the data centers that house them has exploded.

As a result, companies are increasingly turning to partnerships and rental agreements rather than waiting for new facilities to be built. A similar trend was seen recently when Anthropic signed an agreement to use SpaceX's Colossus 1 supercomputer to run its AI models.

Why is SpaceX renting out its AI infrastructure?

The answer lies in a major corporate change. Following SpaceX's merger with Elon Musk's AI company xAI, SpaceX gained control of massive AI infrastructure, including the Colossus data center clusters that were originally built to support xAI's Grok chatbot and other AI projects.

Rather than leaving unused computing power idle, SpaceX has decided to rent out excess capacity. The company said the arrangement gives it flexibility in how it allocates and monetises infrastructure while generating immediate revenue from hardware that xAI is not currently using.

That strategy is proving highly lucrative.

A business worth billions every month

Google's agreement comes just months after Anthropic signed a similar deal with SpaceX. Under that arrangement, Anthropic agreed to pay $1.25 billion per month through 2029 for access to available computing capacity at the Colossus 1 data center near Memphis, Tennessee.

Together, the Google and Anthropic agreements have transformed AI infrastructure into a major business for SpaceX, creating a compute-related revenue run rate exceeding $2 billion per month.

The timing is particularly significant because SpaceX is preparing for what could be one of the biggest IPOs in history, with the company reportedly targeting a valuation of more than $1.75 trillion. Large, recurring infrastructure contracts could make the company even more attractive to investors by providing predictable long-term revenue.

What happens if the capacity isn't delivered?

The agreement includes safeguards for Google. According to the filing, SpaceX must provide the promised computing capacity by September 30, 2026. If it fails to do so, Google can either terminate the agreement after a one-month grace period or accept fewer GPUs and pay reduced fees.

The contract also includes a cancellation clause that allows either company to end the agreement with 90 days' notice after December 31, 2026.

For now, the deal serves as another sign of just how intense the race for AI computing power has become. In the AI era, having the best model matters, but having enough computers to run it may be just as important.

- Ends