SpaceX Signs Up to $6.3 Billion Compute Deal with Reflection AI
N.R. Finch
SpaceX will lease Nvidia GB300 chips at its Colossus 2 data center to open-source AI startup Reflection AI for $150 million a month — a deal worth up to $6.3 billion through 2029, turning Musk's internal AI infrastructure into an external revenue line.
How big is this deal?
Reflection AI will pay SpaceX $150 million per month starting July 1, 2026, for access to Nvidia GB300 chips — Nvidia's latest-generation GPU for training and running AI models — housed in SpaceX's Colossus 2 data center.
If the contract runs through end of 2029, the total reaches roughly $6.3 billion.
But there is an exit clause: after three months, either side can terminate with 90 days' notice. This means → $6.3 billion is a ceiling, not a locked commitment. Both parties are preserving flexibility.
How did Colossus go from internal tool to external business?
The Colossus data center was originally built to power Grok, Musk's AI chatbot positioned as a competitor to ChatGPT.
After a record-breaking IPO, SpaceX expanded its data-center investment and began selling surplus compute externally. Before Reflection, SpaceX had already struck compute deals with Anthropic, Google, and Cursor — and is currently acquiring Cursor.
In plain terms = SpaceX realized it had more top-tier GPUs than it could use internally, so it became a "compute landlord," competing directly with cloud providers and AI infrastructure companies for the same pool of customers.
Who is Reflection AI, and why open source?
Reflection AI is a lab focused on open-source models. It has not yet released a public frontier open-source model, but has gained traction with government and national-security clients — including work on the U.S. Department of Energy's Genesis Mission and broader Pentagon AI programs.
A Reflection spokesperson said: "More and more countries and enterprises are recognizing the risks and costs of relying entirely on closed-source models."
This reflects a broader shift: governments and enterprises are reassessing their dependence on closed-source AI, and open-source models are emerging as a seriously considered alternative.
What does this deal signal for the AI compute market?
Access to advanced Nvidia chips remains one of the biggest bottlenecks for companies training and running frontier models.
By opening Colossus to outside customers, SpaceX has formally entered the race against cloud providers and AI infrastructure companies for scarce GPU capacity.
This means → compute is no longer just a technical resource — it is a standalone commercial asset that can be priced and leased. Whoever holds surplus top-tier GPUs now holds an additional revenue stream.
Content is for reference only, not financial advice.