SpaceX built what was supposed to be its AI moat — the Colossus 1 data center, a massive compute cluster that was central to the company's IPO pitch. There's just one problem: SpaceX's own engineers couldn't make it work for Grok, their in-house model. Latency issues made it impractical.
So now Anthropic gets to rent it instead.
The Infrastructure Play
According to Bloomberg, SpaceX decided to rent out Colossus 1 after internal teams struggled to use it for Grok development. The deal makes sense strategically — SpaceX gets to monetize an underutilized asset, and Anthropic gets badly needed compute capacity at a time when every GPU cluster in America is spoken for.
But there's a deeper signal here. Even the most well-funded company in the world — one that's about to become the largest IPO in history, valued at over $2 trillion — couldn't solve the compute puzzle internally. The bottleneck isn't money. It's expertise in running AI infrastructure at scale.
Even SpaceX couldn't crack the AI infrastructure puzzle internally. The bottleneck isn't money — it's expertise.
What This Means for the Industry
This isn't just a SpaceX story. It's a preview of how AI infrastructure will work going forward: specialized operators who know how to run clusters at scale will own the real estate, while model developers rent rather than build. The days of every big tech company trying to be its own datacenter operator are numbered.
Anthropic just signed 12+ direct data center leases — a first for the startup. Combined with the SpaceX deal, it's clear Anthropic is serious about owning its compute supply chain. Google is potentially providing financial backing.
The irony: SpaceX spent years building the largest compute cluster in Silicon Valley. Now it's a landlord. The rocket company couldn't make Grok work — but it turns out that's a much harder problem than launching rockets.
This deal signals that AI infrastructure is becoming its own category. The companies that can run clusters well will own the platform. Everyone else will rent.