NVDA
SPCX
Chip buffet, anyone?
Elon Musk just threw out a number that sounds fake until you remember we live in the AI era: SpaceXAI’s Colossus 2 could end the year with more than 1.2 million Nvidia GB200 and GB300 processors humming away if the expansion plan holds.
Right now, Musk says the setup already includes 110,000 GB200s and 440,000 GB300s. Then comes the drip-feed of more hardware — 220,000 GB300s next week, another 220,000 in November, and maybe one last 220,000 by late December if the data-center gods cooperate. That’s not a server room. That’s a small city wearing a GPU crown.
Why investors should care
This matters because Nvidia doesn’t just sell chips anymore — it sells the picks and shovels for the AI gold rush. If Colossus 2 keeps scaling like this, it underscores just how sticky demand remains for Blackwell-era hardware.
It also shows the real bottleneck in AI isn’t only silicon. Musk said the weird 110,000-chip increments reflect the number of fiber connections that can actually be plugged into the system. Translation: you can’t brute-force your way past physics. You need power, cooling, networking, permits, and a little bit of luck.
The bigger picture
There’s also a commercial twist here. Colossus is drifting from “internal training beast” to “AI compute for rent,” with reports that Reflection AI and Google have signed giant capacity deals tied to the cluster.
So yes, this is about Nvidia chips. But it’s also about the entire AI supply chain learning the hard way that the next phase of growth looks less like software scaling and more like building a futuristic utility company. Big picture: the winners may be the ones selling the plumbing, not just the apps.