SpaceX awards $6.3B AI compute contract to Reflection AI

SpaceX has finalized a $6.3 billion compute contract with Reflection AI, a two-year-old open-source AI startup founded by former Google DeepMind researchers. The agreement grants Reflection AI direct access to NVIDIA GB300 infrastructure housed at SpaceX’s Colossus 2 data center near Memphis, with operations set to begin in July 2026. The deal spans three years, though either party may terminate it with 90 days’ notice after the initial three-month period.

What the deal entails

The contract allocates $150 million per month to Reflection AI, positioning it alongside larger arrangements SpaceX has reportedly struck with Google and Anthropic. While smaller in scale, the deal underscores the growing trend of AI labs securing dedicated compute infrastructure rather than relying solely on hyperscale cloud providers. The Colossus 2 facility, designed to support high-performance workloads, will host Reflection AI’s operations, reflecting a broader industry shift toward specialized AI infrastructure.

Reflection AI’s focus on open-source models aligns with a rising demand for customizable AI solutions, particularly among enterprises seeking greater control over deployment and data sovereignty. The startup’s access to NVIDIA’s latest hardware could accelerate its development of frontier models, though the contract’s flexibility—allowing either side to exit after the first three months—suggests a cautious approach to long-term commitments amid rapid technological change.

Market implications

The deal highlights two key trends in AI infrastructure procurement. First, the scarcity of high-end GPUs has pushed AI labs to secure direct access to hardware, often bypassing traditional cloud providers. Second, the rise of specialized AI cloud providers—dubbed "neoclouds" by analysts—is reshaping the competitive landscape. Gartner projects these providers could capture 20% of a $267 billion AI cloud market by 2030, driven by enterprise demand for GPU capacity and sovereignty assurances.

For infrastructure buyers, the shift signals a move away from treating memory, storage, and networking as secondary concerns. As AI workloads grow more complex, companies like Micron and Anthropic are collaborating on memory architecture and supply planning, indicating that hardware optimization is now a priority alongside compute procurement. This evolution could lead to more integrated infrastructure designs, where AI labs influence component development rather than relying on off-the-shelf solutions.

What to watch

The contract’s flexibility may set a precedent for future AI infrastructure deals, particularly as startups seek to balance long-term commitments with the need for agility. Observers will also monitor how Reflection AI leverages its access to NVIDIA’s GB300 infrastructure, especially in comparison to larger players like Google and Anthropic. Additionally, the broader adoption of neoclouds could pressure hyperscalers to adapt their offerings, potentially leading to more modular or sovereign-focused AI cloud solutions.