Reflection AI will pay SpaceX $150 million every month starting July 1, 2026, through 2029 for access to Nvidia’s GB300 AI chips and related hardware in the Colossus 2 data center near Memphis. That commitment adds up to as much as $6.3 billion, with either side able to cancel on 90 days’ notice after the first quarter. By comparison, Anthropic and Google each signed much larger deals—$1.25 billion and $920 million per month, respectively—also running through July 2029. The startup launched in 2024, founded by two ex–Google DeepMind engineers, and this is its first major compute agreement. Reflection bills itself as an open-weight AI lab, which means it publishes its trained model parameters instead of keeping them proprietary. After the U.S. blocked Anthropic’s closed models Fable and Mythos, open-weight approaches gained fresh attention. Reflection argues that public models cut down on vendor lock-in and geopolitical risk. SpaceX’s Colossus facility began as xAI’s private training hub before Elon Musk folded it into SpaceX. Once xAI’s own AI plans lost steam, SpaceX opened the racks to outside teams. Now labs like Reflection, Anthropic and Google tap into that idle capacity. For Reflection, landing this deal signals that even a smaller startup can secure top-tier hardware on par with industry giants.