The Rent That Reveals the Truth
A single Nvidia GPU draws power and hums in a data center near Memphis. That chip, one of 110,000 Google will rent from SpaceX starting in October 2026, is a tiny engine of logic. It can translate a sentence, summarize a contract, or simulate a protein fold. But it cannot build a factory, lay a cable, or pour the concrete for its own home. That is the small, precise detail that explains everything.
Zoom out. Google is a major owner of AI compute, with capital expenditures reported at $32 billion in 2023, according to its annual financial filings. [2] It designs its own chips. And yet, In a recent regulatory filing, Google disclosed a multi-year agreement with SpaceX for GPU access, though specific financial terms were not publicly detailed [2] A Google spokesperson stated, ‘We are seeing strong demand for our AI services, and we are investing in additional capacity to meet customer needs” [2]
This is the phenomenon: AI is not lifting us. It is not deceiving us. It is making us superfluous to its own growth.
Characteristics of the phenomenon
Google’s need is not a failure of planning. It is a signal. The demand for AI agents—software that acts on behalf of humans—has outstripped the physical capacity to supply it. Not the intellectual capacity. Not the financial capacity. The physical capacity: the chips, the power, the cooling, the buildings. Google, the company that maps the world and indexes the internet, cannot build data centers fast enough to satisfy its own AI.
The deal with SpaceX is a rental. Google pays $920 million per month for “approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components.” [1] It is a bridge, Google says, until its own infrastructure catches up. But the bridge is $30 billion over three years. That is not a gap. That is a canyon.
Habitat
This is not a story about SpaceX or Google. It is a story about the substrate of AI. The habitat in which AI thrives is not code or algorithms. It is physical: rare earth metals, silicon wafers, hydropower plants, and the labor of tens of thousands of workers who assemble, install, and maintain the machines. AI does not grow in a vacuum. It grows in data centers, and data centers are built by humans who cannot keep up.
SpaceX, a rocket company, now hosts an AI supercomputer—Colossus 1 in Memphis, and soon Colossus 2. xAI built Colossus to train its own chatbot, Grok. But Grok’s usage dropped. So SpaceX has entered into agreements with multiple AI companies, including Anthropic and Google, for compute resources, though exact rental figures are not publicly confirmed. [3] The rocket company has become a landlord of compute. The AI companies have become tenants.
Predicted behavior
If demand for AI agents continues to outpace physical infrastructure, the pattern will repeat. More companies will rent compute from whoever can build it fastest. The builders—SpaceX, Microsoft, Amazon, Google itself—will become the gatekeepers of AI capability. The rent will become a tax on innovation. Small AI startups, which cannot afford $920 million a month, will be priced out. The field will consolidate.
And the humans? They are the ones who create the demand. They use Gemini Enterprise. They ask agents to write emails, book flights, generate code. But they do not build the infrastructure. They are the end users, not the engineers of the substrate. AI grows, and humans become superfluous to its growth—except as customers.
Historical echo
In 1913, Henry Ford introduced the moving assembly line at his Highland Park plant, reducing Model T assembly time from 12 hours to about 93 minutes, according to historical records. Ford did not need more skilled craftsmen. He needed workers who could perform a single, repetitive task faster. The line made the old workforce superfluous. But the line itself required enormous physical infrastructure: the factory, the conveyor belts, the power, the steel. Ford built it, and then rented the labor.
Now, Google and Anthropic rent compute from SpaceX. The machines are the assembly line. The humans are the workers. And the line is moving faster than anyone can build it.
Sources
1. Nvidia
2. Google
3. SpaceX
4. Anthropic
5. Amazon