Open-source artificial intelligence lab buys $6.3 billion in SpaceX computing

Reflection AI, a two year old startup that develops open-weight AI models, has signed a three year $150 million / month ($6.3 billion total) contract with SpaceX, starting July 1st 2026, giving them guaranteed access to the latest Nvidia GB300 chips at the Colossus 2 data center outside of Memphis, Tennessee. This is one of the biggest known open AI infrastructure contracts yet signed.

Founded in 2024 by former Google DeepMind researchers, Reflection’s goal is to develop open-source counterparts to closed-source AI labs. Open-weight AI models are those whose weights are publicly available for download and local use, rather than being served through a cloud API. The contract signing comes shortly after the US government forced Anthropic to disable their strongest frontier models worldwide, cutting off users in dozens of countries from the tools they were developing.

1How big is $150M a month, exactly?

It appears that SpaceX has also signed similar contracts with other major AI companies, such as $1.25 billion / month for Anthropic and $920 million / month for Google. While Reflection’s contract is the smallest of the three, it’s still equal to the entire yearly revenue of a healthy mid-sized tech company, paid each and every month. Together, these contracts show that SpaceX has emerged as a major player in the global AI infrastructure space, raking in billions of dollars from their data centers while the rockets get all the attention.

Each party may cancel the agreement within 90 days upon the expiration of the initial 3-month period, giving both parties an opportunity to back out should the relationship sour. In their press release, Reflection emphasized the importance of open source in light of recent events: “These recent developments underscore the critical role that open source plays in the AI landscape, as more countries and organizations understand the dangers and expenses associated with depending on a few exclusive labs to create fundamental AI abilities.”