SpaceX has been quietly building out its AI infrastructure, and now it’s found a tenant. Reflection AI, an open-source AI lab you might not have heard of yet, signed a deal to rent compute on SpaceX’s Colossus 2 data center near Memphis, Tennessee.
Starting July 1, 2026, Reflection AI will pay $150 million a month for immediate access to Nvidia’s latest GB300 AI chips and supporting hardware. That’s not a typo — $150 million per month, running through 2029.
That’s a staggering number. For context, that’s roughly $1.8 billion a year, which puts this deal in the same ballpark as some of the largest cloud compute agreements we’ve seen. But what makes it interesting is who’s paying and who’s collecting.
Reflection AI isn’t a household name like OpenAI or Anthropic. It’s an open-source-focused lab, which means its research and models are meant to be public. That’s a different philosophy from the proprietary playbooks of most big labs. And they’re betting big on having their own slice of dedicated hardware rather than renting from AWS or Azure.
SpaceX, meanwhile, is already running Colossus 2 for its own AI work — things like Starlink optimization, autonomous drone control, and whatever else Elon’s empire needs. This deal suggests they have enough spare capacity (or are willing to carve out a dedicated partition) to justify a multi-year, multi-billion dollar lease.
I’m curious about the actual hardware setup. The GB300 is Nvidia’s latest, and it’s not exactly sitting on shelves. SpaceX must have secured a serious allocation to be able to offer guaranteed access. The fact that Reflection AI gets “immediate access” starting July 1 tells me the infrastructure is already built and running. No waiting for construction.
The location matters too. Memphis isn’t Silicon Valley or Northern Virginia. It’s a strategic pick for power availability and cooling — the Tennessee Valley Authority has plenty of juice, and the climate is milder than Texas or Arizona for those hot-running chips. Plus, it’s close to major fiber routes.
What I don’t know is whether this deal includes any equity component or if it’s purely cash. $150 million a month is a lot for a research lab, even one with serious backers. Reflection AI must have either deep pockets or a plan to monetize their work quickly. Open-source labs typically don’t have that kind of burn rate unless they’re backed by a big patron or have a commercial arm.
This also raises questions about SpaceX’s long-term AI strategy. Are they building a cloud business? Or is this just a one-off lease to offset their own infrastructure costs? My guess is the latter — SpaceX isn’t in the cloud hosting game, but if you’ve got a giant cluster running anyway, selling access makes financial sense.
For the open-source AI community, this is a win. Reflection AI gets dedicated compute without the cloud vendor lock-in that comes with AWS or Google Cloud. They can run experiments, train models, and release them freely. That’s increasingly rare as compute costs skyrocket.
Will this model spread? I doubt it. Most open-source labs can’t afford $150 million a month. But it signals that there’s room for alternative compute arrangements outside the big three cloud providers. If you’ve got the hardware and the right partner, you can make a deal that works for both sides.
I’ll be watching to see what Reflection AI produces with this access. If they deliver something meaningful, this deal will look like a bargain. If not, it’ll be a very expensive lesson in how hard it is to compete with the proprietary labs.
Comments (0)
Login Log in to comment.
Be the first to comment!