SpaceX sells $6.3B in Colossus compute to Reflection AI
The deal positions Musk's space company as a major AI infrastructure provider while backing open-source alternatives to closed models.
SpaceX has secured a major customer for its Colossus data center infrastructure, signing an agreement with open-source AI startup Reflection AI worth potentially $6.3 billion over three years.
Under the terms, Reflection AI will pay SpaceX $150 million monthly starting July 1, 2026, running through 2029. In return, the startup gains immediate access to Nvidia GB300 chips—among the most advanced processors available for training and deploying AI models. Either party can terminate the contract with 90 days' notice after an initial three-month period.
Why it matters
The agreement signals SpaceX's evolution beyond aerospace into AI infrastructure sales, positioning the company to compete with cloud providers in the race to monetize scarce GPU capacity. It also arrives as enterprises and governments reconsider their dependence on closed AI systems following recent access disruptions, potentially accelerating adoption of open-source alternatives that offer greater control and transparency.
Colossus becomes a compute marketplace
SpaceX originally built its Colossus facility in Memphis to power Grok, Elon Musk's AI chatbot designed to rival ChatGPT. Now the company is leveraging that infrastructure to sell computing capacity to external customers, joining a roster that already includes Anthropic, Google, and Cursor—the latter of which SpaceX is acquiring.
The strategy transforms what Musk has called a "gigafactory of compute" into a revenue-generating asset that helps justify SpaceX's expanding footprint beyond rockets and Starlink satellite internet. Investors have been monitoring whether the company can successfully diversify into AI services and data center operations.
Open-source momentum
Reflection AI represents a strategically distinct customer: an AI lab focused exclusively on open-source models at a time when the risks of closed systems have become more visible. The startup, most recently valued at $25 billion, is building what it calls "American open intelligence"—frontier models that governments and enterprises can inspect, customize, and operate with greater autonomy than proprietary alternatives from OpenAI, Anthropic, or Google.
The company has not yet released a public frontier model but has gained traction with national security customers, including work with the Department of Energy's Genesis Mission and Pentagon AI initiatives.
"Recent events highlight how important open source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models," a Reflection spokesperson said.
The statement references recent disruptions when Anthropic cut off access to certain services, underscoring vulnerabilities in relying on closed-model providers for critical operations.
Compute as strategic currency
Access to advanced Nvidia chips remains one of the most significant bottlenecks for companies attempting to train and serve frontier AI models. By opening Colossus to outside customers, SpaceX positions itself alongside major cloud providers racing to sell limited GPU capacity in a market where compute itself has become strategic currency in the AI competition.
The financial details were first reported by CNBC, which reviewed materials outlining the agreement's terms.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call