SpaceX isn’t just a space company anymore — it’s becoming one of the world’s largest AI infrastructure players. Starting July 1, SpaceX will lease Nvidia GB300 chips from its Colossus 2 data center in Memphis to AI startup Reflection AI. The price tag: $150 million per month.
The numbers
The contract runs for 42 months with a 90-day exit clause. Total value: $6.3 billion. This pushes SpaceX’s contracted AI compute revenue past $2.3 billion per month — a figure that should make even the big cloud providers take notice.
Reflection AI, a startup focused on efficient inference and model optimization, gets access to Nvidia’s latest chips. GB300 is Nvidia’s current datacenter GPU generation — even more powerful than the H100 and B200 chips that are already scarce.
Why SpaceX?
SpaceX has an advantage traditional cloud providers don’t: Starlink infrastructure. The combination of its own fiber, data centers, and global network makes SpaceX a serious competitor in the compute market.
What’s happening here is remarkable: a space company is becoming a GPU landlord. Elon Musk has xAI and Grok as his AI bet — but leasing compute through SpaceX could end up generating far more revenue than his own AI model.
What this means for the market
For startups like Reflection AI, this deal is a game-changer. GPU access is the biggest bottleneck in the AI industry, and anyone who can pay $150 million per month gets hardware that’s otherwise reserved for Big Tech.
For the major cloud providers — AWS, Azure, Google Cloud — SpaceX is an unexpected new competitor. With over $2.3 billion monthly in compute contracts, SpaceX is suddenly playing in the same league as the established providers.
Sources: