An Intellyx Brain Candy Update from KubeCon NA 2025
Nebius was at KubeCon NA 2025 demonstrating its hyperscaled globally distributed cloud platform purpose-built on Kubernetes with KubeVirt clusters and software-defined networking for hosting and elastically scaling AI training, inference and agentic workloads.
AI project teams would rather focus on improving parameters and building model behavior than opening up a box of tools for provisioning GPUs and TPUs, launching clusters, defining networks and wrangling costly object storage volumes. Nebius abstracts away the complexity of selecting and operating an ideal AI infrastructure with a no-code-required provisioning interface, fast failover recovery, and 24/7 human support.
“It’s not at all like the experience you get with a traditional hyperscaler. Deployment speed and performance matters, but what you really need is high-speed recovery, so that if one of your GPUs inevitably fails during model training, you don’t lose everything and have to start the whole Slurm job over,” said Aleksandr Patrushev, head of ML/AI products, Nebius.
Part of the secret sauce of Nebius is the way their software-defined infrastructure and virtual clusters can ‘pass through’ the global data center capacity of their GPUs and commodity hardware, allowing workloads to run and maintain state wherever best suited, separate from the customer-level concerns of tuning for token costs or performance objectives.
Copyright ©2025 Intellyx B.V. Intellyx is an industry analysis and advisory firm focused on enterprise digital transformation. Covering every angle of enterprise IT from mainframes to artificial intelligence, our broad focus across technologies allows business executives and IT professionals to connect the dots among disruptive trends. At the time of writing, Nebius is a former Intellyx customer. No AI was used to write this article. To be considered for a Brain Candy article or event visit, email us at pr@intellyx.com.
