SpaceX has added another major AI customer to its expanding compute business. Open-source startup Reflection AI will begin paying $150 million per month from July 1, 2026, for access to Nvidia's latest GB300 chips and related hardware at SpaceX's Colossus 2 data center near Memphis, Tennessee.
The agreement runs through 2029 and could reach a total value of up to $6.3 billion. Either side may exit the contract with 90 days' notice after the first three months, giving the deal flexibility while still marking a significant infrastructure commitment.
Reflection AI described the arrangement as a milestone for its open-weight strategy, which releases trained model parameters publicly and positions the company as an alternative to closed AI labs. Founded in 2024 by former Google DeepMind researchers, the startup said the partnership strengthens its ability to scale open models for broader use across research and industry.
The Colossus data center, originally developed by xAI, has become a key asset in SpaceX's broader AI infrastructure play. As demand for advanced chips continues to rise, the site is emerging as a powerful platform for frontier model development and large-scale experimentation.
This deal reflects how compute access is becoming a defining layer of the AI economy, shaping who can build, train, and expand the next generation of intelligent systems.