SpaceX Rents Out Its Supercomputers to Open-Source AI (for $150M a Month)

Here's the jaw-dropper: Reflection AI is paying $150 million per month — not per year, per month — from July 2026 through the end of 2029 for priority access to Nvidia's latest GB300 chips and the full supporting infrastructure inside Colossus 2, SpaceX's data center near Memphis, Tennessee. Yep, you read that right. Nearly two billion dollars over three years to play with cutting-edge hardware.

To put that in perspective: usually tech companies join waiting lists for weeks just to get their hands on Nvidia's latest chips. Reflection AI decided to skip the line by buying the right to use them immediately. It's like paying to cut in line at the most exclusive concert in the history of AI.

Here's the twist: Reflection AI is an open-source AI lab — basically an organization that builds AI models anyone can use for free. This isn't some Silicon Valley startup trying to become the next OpenAI. It's a team that wants to speed up AI research with publicly accessible resources. The irony is thick: they're spending a fortune on compute they'll probably share with everyone else.

From SpaceX's angle, this is pretty smart business. Elon Musk has always understood that the real value isn't in launching rockets or Tesla batteries — it's in data and compute. Renting out excess computing capacity is like discovering your factory produces a byproduct that people are willing to pay top dollar for. Colossus 2 was built to handle SpaceX's internal operations, but they clearly figured out there was serious money in renting the rest.

The timing is interesting too, because compute and chip costs have become the new bottleneck in AI. It's not about algorithms anymore — it's about money and hardware. Whoever has access to compute has power. And now Reflection AI just paid to not wait in line.