Design · Integration · 24×7 support

Supermicro servers from Spain.

We design, integrate and support Supermicro platforms on RHEL, SUSE or Ubuntu for Ceph, OpenStack, virtualisation, AI and HPC. No closed catalogues: every platform is sized from the actual workload.

Why on-prem in 2026

Three reasons to look
at Supermicro now.

×5-10 VMware post-Broadcom Broadcom pushed vSphere licences 5× to 10× higher on most enterprise contracts renewed through 2024-25. The way out runs through OpenStack, OpenShift Virtualization or Proxmox — and all of them need validated hardware underneath.
10 kW HGX H200 GPU node A chassis with 8× H200 GPUs reshapes the whole project: cooling, PDU sizing and 400 GbE networking stop being optional.
90 bays SuperStorage 4U Ceph on commodity hardware replaces NetApp or Dell PowerStore. Petabytes at a fraction of the cost when the cluster is designed well.
Workloads we run on Supermicro

Four use cases
we own end to end.

Every project starts the same way: which workload, at what scale, with which SLA. From there we choose the Supermicro family, the network, the storage and the Linux stack. These four are the ones that come through the door most often.

GPU SuperServer · AI/ML

Artificial Intelligence

Training and on-prem LLM fine-tuning, inference at scale, and RAG on data that cannot leave your network. The realistic alternative to renting hyperscaler GPUs once the project runs longer than a year.

  • Architecture4 to 16 GPU nodes + head + shared NVMe storage
  • HardwareGPU SuperServer 4U/8U · 8× HGX H100/H200 or AMD MI300
  • NetworkInfiniBand NDR 400 Gb/s for inter-node all-reduce
  • StackOpenShift AI, RHEL AI or Kubernetes + NVIDIA GPU Operator
SuperBlade + TwinPro · HPC

HPC · Heavy computation

Scientific clusters, CFD, numerical simulation, genomics or risk engines. Thousands of cores, low-latency fabric and a scheduler that squeezes every watt from the rack.

  • ArchitectureHead + login + homogeneous compute + parallel storage
  • HardwareSuperBlade 8U (10-20 blades) or TwinPro 2U · EPYC Turin
  • NetworkInfiniBand NDR or RoCEv2 over 100/400 GbE
  • StackSlurm + OpenHPC, Rocky Linux or RHEL, Lustre or BeeGFS
SuperStorage · Ceph SDS

Ceph SDS · Storage

Replace NetApp or Dell PowerStore arrays with your own Ceph cluster: block, S3 object and file from the same hardware. PB-scale backup without per-TB licensing.

  • ArchitectureMinimum 5 OSD + 3 MON, 3× replication or erasure coding
  • HardwareSuperStorage 4U · 45-90 bays · NVMe WAL/DB + HDD tier
  • Network2× 25/100 GbE separating public and cluster network
  • StackCeph Reef on RHEL or SUSE, or Rook inside OpenShift
SuperServer + SuperBlade · OpenStack

OpenStack · Private cloud

The realistic way out of VMware post-Broadcom when Proxmox is not enough: an AWS/GCP-style private cloud in your own datacentre, with multi-tenant, quotas, images and API. Bare-metal, VMs and containers on the same platform.

  • Architecture3 HA controllers + N compute + Ceph as Cinder backend
  • HardwareSuperServer 1U/2U (control + compute) or SuperBlade for density
  • Network2× 25 GbE minimum · OVN SDN · separate management network
  • StackRed Hat OpenStack Services on OpenShift (RHOSO), Canonical or SUSE
How we deliver the project

Three phases
and no surprises.

01 Design and proposal Technical session with your team to scope workload, scale and SLA. We come back with the architecture, the shortlist of Supermicro models and a budget broken down into concrete line items. discovery · sizing · BOM
02 Procurement, racking and integration Order through the authorised channel, rack mounting, BMC/IPMI provisioning and Ansible-based deployment on RHEL, SUSE or Ubuntu. On top of that Ceph, OpenStack, OpenShift or Kubernetes — whichever the case calls for. channel · burn-in · ansible
03 24×7 support and evolution 24×7 on-call from the same people who designed the platform. Periodic healthchecks, evolutive maintenance and training for your in-house team so they don't always depend on us. 24×7 · healthcheck · SRE
Technical session · 30 min

Tell us your workload.
We'll send back an architecture.

30-minute technical session, no strings attached. Tell us what you plan to build (Ceph, OpenStack, HPC, AI, virtualisation) and at what scale. We come back with a sketched architecture, the shortlist of Supermicro models and a rough effort estimate. No sales talk.

We are independent. No conflicts of interest with any vendor and no commissions for pushing one over another. We recommend what suits you, not what pays us.

SIXE