SpaceX has signed a compute deal worth up to $6.3 billion with Reflection AI, an open-source AI startup aiming to build frontier models that compete with OpenAI, Anthropic, and Google. The agreement, reported by CNBC on June 22, 2026, sees Reflection paying $150 million per month starting July 1 for access to Nvidia GB300 chips housed at SpaceX’s Colossus 2 data center in Memphis, Tennessee.

The deal runs through approximately 2029, though either party can exit with 90 days notice after the first three months. Reflection AI, last valued at $25 billion, has Nvidia among its backers — with the chip maker investing $800 million in the startup. That means Nvidia is effectively on both sides of this arrangement: supplying the hardware SpaceX rents out and investing in the company renting it.

SpaceX Is Becoming an AI Infrastructure Company

The Reflection deal is the latest evidence that SpaceX has turned Colossus from an internal compute asset into a commercial platform for AI development. The roster of tenants now includes Anthropic, which is paying roughly $1.25 billion per month, Google at approximately $920 million per month, and now Reflection at $150 million per month. Cursor, the AI coding tool, was acquired by SpaceX outright.

That is a meaningful business line building alongside SpaceX’s rocket and satellite operations. The Colossus 2 facility in Memphis is positioned as one of the largest concentrations of Nvidia’s newest GPU architecture, and the willingness of AI labs to pay at this scale reflects how central access to raw compute has become to staying competitive at the frontier.

Who Is Reflection AI

Reflection was founded by two former Google DeepMind researchers and has focused on what it describes as “American open intelligence” — frontier-class models that governments, enterprises, and developers can use without the access restrictions that come with closed systems. The company has not yet released a public frontier open-source model, but has been building momentum with government and national security customers.

The $6.3 billion commitment signals serious intent. Frontier AI model training at the scale required to compete with GPT-5.5 or Claude Opus 4.8 demands sustained access to clusters numbering in the hundreds of thousands of GPUs. The Colossus 2 deal gives Reflection the infrastructure to attempt that training run.

What This Means for Business

Compute is still the scarce resource. Despite years of predictions that model training costs would fall, the leading AI labs are signing deals measured in billions of dollars per year. The implication for businesses is that access to frontier AI capabilities will remain concentrated at the provider level for the foreseeable future — which is why your choice of AI vendor matters more than it might appear.

Open-source AI is getting serious backing. Reflection’s model is to build powerful AI that organizations can deploy without dependency on a single provider. If the company succeeds, it expands the options available to businesses that cannot or do not want to route sensitive workflows through closed, US-based commercial APIs. Enterprise teams with data sovereignty concerns or government contract constraints should watch Reflection’s progress closely.

The AI infrastructure layer is consolidating around a small number of physical sites. SpaceX’s Colossus, CoreWeave facilities, and dedicated hyperscaler zones are where frontier AI is being built. This physical concentration is becoming a strategic consideration for companies thinking about AI supply chain risk — including what happens if a regulatory action, a natural disaster, or a contractual dispute affects a facility where their AI provider trains its models.

Nvidia wins regardless of who wins the model race. Nvidia’s investment in Reflection is a hedge against its own business risk: if open-source AI grows and commoditizes the model layer, Nvidia wants exposure to that upside, not just the enterprise closed-model market. For businesses evaluating their AI tech stack, this is a reminder that the infrastructure economics favor diversification across multiple model providers.

The Bigger Picture

The compute deals piling up at Colossus 2 are reshaping what we think of as AI companies. SpaceX was a rocket company. Colossus started as an internal compute asset for Grok and xAI. It is now generating what may become tens of billions in annual infrastructure revenue from the leading AI labs.

That evolution matters for anyone building a business around AI: the economics of the underlying compute layer are not settled, and the companies controlling that layer are not who most people expected. As the frontier model race continues to accelerate, access to physical compute — not just model capability — will remain the key variable.

---

Enterprise DNA works with businesses to identify the right AI capabilities for their situation and build the internal skills to use them effectively. Book a discovery call to talk about what AI infrastructure choices mean for your business.