There’s something fitting about an AI cloud company needing AI just to keep its own cloud running smoothly. Nebius builds hyperscale cloud environments for training and running AI models, and it just picked Komodor‘s autonomous SRE platform to handle troubleshooting across its GPU-heavy cloud infrastructure. That choice reveals just how complicated these systems have gotten.
Running a fleet of interconnected servers built for AI training isn’t like managing a typical data center. Nebius relies on custom GPU scheduling layers, along with Kubernetes tooling stacked with extensions that go well beyond what comes out of the box in standard cloud setups. That customization makes the platform powerful, but it also means a small misconfiguration somewhere can ripple into a much bigger failure. As a result, figuring out exactly where things went wrong by manually digging through cloud logs and dashboards simply doesn’t scale anymore.
“Nebius operates AI cloud infrastructure at scale,” said Danila Shtan, the company’s CTO. “Uptime and performance are mission-critical, and require fast, well-grounded incident investigation across complex Kubernetes environments.” That is essentially the job Komodor exists to do.
Instead of just flagging anomalies the way older cloud monitoring tools do, Komodor’s platform, powered by something it calls Klaudia, actually investigates incidents on its own. Multiple specialized agents handle different domains across the cloud stack, including networking, storage, and GPUs, running commands and pulling logs to trace a problem back to its actual cause rather than leaving that work for an engineer at 2am.
Komodor’s co-founder Itiel Shwartz put the underlying problem plainly: as AI workloads pile complexity onto already-stretched SRE teams, manually managing cloud reliability and cost simply stops being sustainable. According to the company, its platform can cut resolution times by 60 to 80 percent in environments like Nebius’s.
What’s notable here isn’t really the specific numbers, since outside parties haven’t independently verified those figures. Instead, it’s the broader pattern. Companies that build cloud infrastructure for AI are increasingly leaning on AI just to keep that same cloud stable, which says something about how fast operational complexity is outpacing what human teams can reasonably monitor by hand.
