Elon Musk’s SpaceX is rapidly evolving from an aerospace pioneer into a critical infrastructure provider for the global artificial intelligence race. A landmark agreement with Reflection AI, an open-source startup, has positioned SpaceX as a formidable landlord in the high-performance computing market. The deal, valued at up to $6.3 billion, involves the leasing of Musk’s 'Colossus' compute clusters to external parties, signaling a strategic diversification of SpaceX’s revenue streams beyond satellite internet and orbital launches.
Under the terms of the agreement, Reflection AI will gain immediate access to NVIDIA’s GB300 chips—the high-end successors to the Blackwell architecture designed specifically for training frontier-scale large language models. Starting July 2026, the startup will pay a monthly rental fee of $150 million for compute capacity. This massive capital commitment highlights the extreme costs and hardware scarcity currently defining the frontier of AI development, where access to the latest silicon is the ultimate competitive advantage.
The partnership underscores the unique vertical integration of the Musk ecosystem, where SpaceX’s energy and cooling infrastructure are repurposed to support massive GPU arrays. While the contract extends through 2029, it includes a flexible exit clause allowing either party to terminate the arrangement with 90 days' notice after an initial three-month period. This suggests a volatile and fast-moving market where compute needs and hardware availability are in constant flux.
For the broader AI industry, this deal validates the growing importance of independent compute providers. As traditional cloud giants like Microsoft and Google prioritize their internal AI projects, SpaceX is carving out a niche as a neutral, high-capacity alternative for open-source developers. The move effectively turns SpaceX into a vital utility for the AI era, ensuring that even 'non-Big Tech' players can access the industrial-scale compute required to challenge the current leaders of the field.