SpaceX has struck a $6.3 billion compute agreement with Reflection AI, an open-source AI lab, according to TechCrunch. Under the deal, Reflection will pay SpaceX $150 million a month from July 1, 2026 through 2029 for access to Nvidia GB300 chips housed at Colossus 2, the data center near Memphis, Tennessee. Either party can walk after the first three months with 90 days' notice.
The arrangement is notable both for its size and its counterparty. Reflection AI was founded in 2024 by former Google DeepMind researchers and has raised roughly $2 billion to build open-weight models -- publicly releasing trained parameters as an alternative to closed frontier labs like Anthropic and OpenAI. That an open-source lab can underwrite a multibillion-dollar, multi-year compute contract is itself a statement about where capital and conviction are flowing.
“The arrangement is notable both for its size and its counterparty.”
For SpaceX, the deal is the latest sign that its data-center capacity has become a serious business line. The Reflection contract is smaller than SpaceX's concurrent agreements with Anthropic ($1.25 billion a month) and Google ($920 million a month), but together the three turn Colossus 2 into one of the most commercially valuable compute hubs in the country. Elon Musk's empire is increasingly a landlord to the AI labs racing to train ever-larger models.
The structure matters as much as the headline number. These are long-dated, recurring-revenue contracts -- closer to infrastructure leases than hardware sales -- which is exactly the kind of cash flow that makes SpaceX's own pre-IPO story more compelling. But the three-month exit clause shows that even at this scale, compute buyers are unwilling to be fully locked in while chip generations and model economics shift underneath them.
The broader signal is that the binding constraint in AI is no longer talent or even capital -- it is access to power and GPUs at scale. As labs sign nine- and ten-figure monthly contracts to secure capacity years in advance, the companies that own the physical buildout are capturing a structurally advantaged slice of the AI economy.