Cedana

Cedana, a Y Combinator S23 alumnus, empowers organizations with hyperscaler-grade orchestration for their AI workflows. At its core, Cedana offers live migration capabilities for both CPU and GPU workloads, leading to significant cost reductions—up to 80%. This technology also accelerates the time to first token by a factor of 2 to 10 and ensures the stateful reliability of training jobs, even in the face of severe GPU failures. Cedana's solution seamlessly integrates with Kubernetes (K8s) and supports Kueue and Slurm for distributed training jobs, as well as Kserve for inference serving.

While internal versions of these capabilities exist within tech giants like OpenAI, Meta, and Microsoft, Cedana aims to democratize access, making them available to a broader audience. The company's overarching vision is to transform cloud compute into a real-time, arbitraged commodity, optimizing resource utilization and efficiency for all users.