AI startup Reflection says it has signed a $1B+ deal to secure computing capacity from Nebius, including access to Nvidia chips, following a deal with SpaceX
Context & Ripple Effects
Reflection previously committed to a large SpaceX computing arrangement for Nvidia GB300 access through 2029. Its new Nebius agreement adds another major capacity relationship as it develops open-source models intended to compete with leading closed-model systems.
Nebius has also been building beyond raw cloud capacity: it agreed to acquire Eigen AI, whose software focuses on optimizing chips for inference. Meanwhile, Microsoft’s previously disclosed long-term Nebius commitment shows that the provider is already serving large-scale AI-compute demand.
First-order effects
- Reflection gains an additional source of Nvidia-based compute capacity under a deal it says exceeds $1 billion, reducing reliance on a single infrastructure relationship as its model-development needs expand.
- Nebius secures a sizable AI-startup customer and further validates its position as a supplier of specialized AI cloud capacity.
Second-order effects
- The agreement increases pressure on AI-cloud providers to lock in long-duration capacity commitments from model developers, rather than compete solely for short-term usage workloads.
- Reflection’s use of both SpaceX and Nebius suggests that large model builders may spread procurement across providers, making dependable chip access and deployment capacity central competitive differentiators for cloud suppliers.
Third-order effects
- If similar arrangements persist, AI infrastructure could increasingly be organized around a small number of capacity-rich providers and multi-year reservations by model companies, rather than purely on-demand cloud purchasing.
- Nebius’s combination of capacity supply and inference-optimization software points toward competition over the efficiency of serving AI workloads, not just access to Nvidia hardware.
The trend: The deal is one data point in the shift toward multi-provider, long-term compute procurement as AI developers seek to secure scarce high-end accelerator capacity at scale.