SpaceX Sells $6.3B in Compute to Open-Source AI Lab
Reflection AI will pay SpaceX $150 million a month — $6.3 billion total — for Nvidia GB300 chips at Colossus 2, joining Anthropic and Google as major SpaceX compute customers.
Evgenii Arsentev · PhDReflection AI, a two-year-old startup building open-weight AI models, will pay SpaceX $150 million a month for three years starting July 1, 2026, securing access to Nvidia's latest GB300 chips at the Colossus 2 data center near Memphis, Tennessee. The total commitment reaches $6.3 billion — one of the largest announced open AI infrastructure deals in the industry so far.
The company was founded in 2024 by researchers who left Google DeepMind, with a stated mission to build open alternatives to closed-source AI labs. Open-weight models publish their trained parameters for anyone to download and run on their own hardware, rather than locking everything behind a cloud API. The timing of the announcement is pointed: it arrives right after the US government ordered Anthropic to shut off its most powerful frontier models globally, leaving users in dozens of countries without access to tools they had built on.
How big is $150M a month, exactly?
SpaceX has quietly signed similar compute-supply arrangements with other AI giants: Anthropic at $1.25 billion per month and Google at $920 million per month. Reflection's $150 million is the smallest of the three — but still equivalent to the annual revenue of a solid mid-size tech company, spent every single month. Taken together, these deals suggest SpaceX has become one of the most significant AI infrastructure players on the planet, generating billions in revenue from its data centers while its rockets make headlines.
Either party can terminate the arrangement with 90 days' notice after the first three months, so both sides have an exit if things don't work out. Reflection pointed to the geopolitical dimension in its statement: "Recent events highlight how important open source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs" of relying on a handful of proprietary labs for foundational AI capability.
If your projects are currently built entirely on proprietary AI, this is a useful prompt to think about a plan B. Reflection's models will be public and runnable on your own hardware — meaning that when the next access disruption hits, there may be a real open alternative ready to use. Worth watching what they actually ship, and whether the open model quality holds up to the closed-source frontier.
Related guides

Author
Evgenii Arsentev
PhD · Chief Product Officer at a tech company
Want to actually build this?
Guides explain. The free course transforms — personalized, gamified, and built to get you shipping fast.
◉ Start the free courseSource: techcrunch.com