Case Study: SpAItial achieves 97% scaling efficiency with Crusoe Cloud

SpAItial hits 97% scaling efficiency with Crusoe Cloud

SpAItial, a company developing physically grounded AI world models, faced a challenge scaling its compute infrastructure for training frontier AI models. The training was compute-bound and required a distributed cluster of high-performance GPUs with a fast interconnect like InfiniBand, which they found difficult to obtain from on-demand cloud providers. This infrastructure constraint was a major bottleneck for their engineering team's rapid time-to-market goals.

By choosing Crusoe Cloud, SpAItial implemented a solution using Crusoe's vertically-integrated AI infrastructure, which included NVIDIA Hopper GPUs and a managed private VPC over NVIDIA Quantum InfiniBand. The use of Crusoe Managed Kubernetes further offloaded the control plane management. As a result, SpAItial achieved a 97% scaling efficiency, reduced the time from initial contact to proof-of-concept to just 11 days, and increased the speed of its training experiments by 30%, allowing its team to focus on core AI research instead of infrastructure management.

SpAItial