Selectel adds flagship servers for model training and expands its image catalog

Selectel has published its February roundup, showing how it is reshaping its infrastructure around growing demand for AI workloads. The main updates are new dedicated servers for training large models, fresh images in the AI marketplace, and several notable changes to its cloud Kubernetes platform.

Servers for training

The headline addition is the GL8-B300-HGX-25GE dedicated server, built on the NVIDIA HGX B300 platform. Selectel calls it the most powerful configuration in its current lineup, designed for AI training, fine-tuning, and production inference of large language models. The company specifically highlights the 2.3 TB of total video memory. That is the level you need not for pilots but for heavy pipelines, long context, and large batches, where the bottleneck is no longer just the GPU but the stability of the entire infrastructure.

"NVIDIA HGX B300 is a flagship platform for large-scale AI workloads."

For teams that don't need top-tier cluster capacity, the company also added more down-to-earth options. The AR45G-NVMe-H200 configuration is positioned as a compact, more affordable server for inference, prototyping, and small-scale training. Another option, the AR45G-NVMe-RTX6000, targets scenarios where 96 GB of VRAM is enough: LLM pilots, computer vision tasks, graphics, and applied workloads. Selectel also expanded its storage lineup with 28 TB HDDs for archives, backups, and logs.

Marketplace and cloud

Selectel's AI marketplace now offers new ready-made images that cover different stages of the ML lifecycle, from experimentation to quality control and data labeling. This isn't catalog padding for its own sake: the company is adding tools that help build a more complete pipeline around a model, rather than just running inference in a container. For engineering teams, it's also a faster way to stand up a working environment without extra manual integration.

• Aim — for tracking experiments and comparing model tuning results.
• Lobe Chat — an interface for working with LLMs, with a focus on plugins and ready-made community solutions.
• Evidently AI — monitoring of model quality and metrics, which can be used to automate updates.
• Xtreme1 — a data labeling service, including lidar and object classification scenarios.

Separately, Selectel brought L4 GPUs with 24 GB of memory to its cloud. The company describes the L4 as a general-purpose card for AI/ML tasks, video processing, streaming, and VDI, as well as applied scenarios such as audio-to-text transcription. The launch matters beyond AI teams: the L4 is a common pick where you need a balance of price, energy efficiency, and adequate performance without moving up to expensive training-grade accelerators. For teams on a tight budget, it's an especially practical option.

Kubernetes and the system layer

A significant infrastructure update landed in Managed Kubernetes. Selectel added full support for the ephemeralStorage resource in Cluster Autoscaler and Karpenter. The practical upshot is simple: when scaling a cluster, the system now more accurately determines how much local ephemeral storage new Pods actually need. Previously, in clusters with no nodes that had ephemeralStorage explicitly set, autoscaling could miscalculate requirements. For teams running data-processing and AI workloads, this eliminates an annoying class of errors during scale-up.

Several other changes cover control and the base system layer. Managed Kubernetes now has audit logs for key operations on clusters and node groups, which simplifies change review, security, and compliance. The interface also now shows private DNS configuration, and the User Data display for node groups has been updated. At the OS level, Selectel released SELECTOS 1.3, which updates packages to Debian 12.13, patches 177 vulnerabilities, and adds a container image. The company also launched Astra Linux for A-DC servers, a move aimed at customers who need a certified, secure environment.

What it means

The February update package shows that Selectel wants to be more than a hardware rental service — it is aiming to be a full-fledged AI infrastructure platform. At the top of the stack, the company is expanding its catalog of ready-made ML tools and cloud GPUs; at the bottom, it is strengthening Kubernetes, base operating systems, and the compliance layer. For teams building their own AI services, this cuts down on manual assembly and shortens the path from pilot to production.