SpaceX just locked in $6.3 billion from an AI startup — and it has nothing to do with rockets. Reflection AI, a $25 billion open-source startup founded by former Google DeepMind researchers, signed a $150 million-per-month compute lease with SpaceX for access to Nvidia GB300 chips at the Colossus 2 data center in Memphis, Tennessee. The deal runs from July 1, 2026 through the end of 2029, and it pushes SpaceX’s total committed AI compute revenue past $80 billion — making it one of the most important AI infrastructure companies in the world.
The agreement was reported by CNBC and confirmed by multiple outlets on June 22. Reflection AI joins Anthropic, Google, and Cursor as tenants at Colossus 2 — a facility SpaceX describes as one of the largest third-party compute platforms on earth. A 90-day exit clause after the first three months gives Reflection an out, but at $150 million a month, SpaceX is already building around this revenue stream.
Reflection AI: The $25B Open-Source Bet Few People Saw Coming
Few people outside AI research circles have heard of Reflection AI, but its pedigree is hard to ignore. The company was founded by two former Google DeepMind researchers and focuses on open-source foundation models — positioning itself as an alternative to the closed ecosystems of OpenAI and Anthropic. Backed by Nvidia with ties to Pentagon and Department of Energy AI programs, Reflection is one of the only startups that can sign a $6.3 billion compute lease and have it make economic sense within months of launch.
Why Colossus 2 Instead of AWS or Azure
The move to Colossus 2 over a traditional cloud provider is deliberate. AWS, Azure, and Google Cloud add abstraction layers and variable pricing on top of the raw silicon — at Colossus 2, Reflection gets direct access to the Nvidia GB300 cluster with fixed monthly rates. For a startup burning through compute on pre-training runs, predictable costs over a multi-year roadmap matter as much as raw performance. SpaceX’s fixed-rate leasing model, unusual in the cloud world, is becoming a compelling proposition for labs that need certainty.
SpaceX Is Quietly Running the AI Internet
The larger story here is what SpaceX has become in 18 months. The company behind reusable rockets is now one of the world’s biggest AI compute landlords, with more than $80 billion in committed revenue from outside AI clients through 2029. That roster now includes Anthropic, Google, Cursor, and Reflection — essentially a who’s-who of frontier AI. If you want to train a state-of-the-art model in 2026 or 2027, there is a real chance the compute lives in a SpaceX data center in Memphis, Tennessee.
Our Take
SpaceX turning from launch provider into AI infrastructure landlord is one of the stranger pivots in modern tech — and it is happening faster than most people realise. The $80 billion in committed compute revenue now rivals what hyperscalers like Microsoft and Amazon have publicly announced in AI buildout. What Colossus 2 makes clear is that AI compute has become infrastructure in the same way electricity is infrastructure: whoever controls the physical hardware controls a chokepoint in the AI economy. For every AI lab trying to scale in 2026, the question is no longer just “which model?” — it is “whose hardware, and can you actually get access to it?”
