Home / Case Studies / Sovereign Cloud Infrastructure
Case Study · Sovereign Cloud · Private Infrastructure

Private cloud for a sovereign operator — 40 hosts, 400 VMs, 14 weeks

A national sovereign cloud operator needed to move government workloads off foreign hyperscaler infrastructure. Haink designed and delivered the full three-layer stack: compute nodes, 100GbE leaf-spine fabric, and an all-flash shared storage pool. First VMs were live on week 14.

Architecture — three layers, one supplier

Haink supplied all hardware layers under a single contract. No coordination overhead between compute, network, and storage vendors.

Layer 1Compute
40× dual-socket compute hosts

2-socket AMD EPYC 9654 (192 vCPUs per host), 768 GB DDR5 ECC RAM, dual 25 GbE server NICs. Hyper-converged-ready: NVMe local cache per host for hot data. VMware vSphere + vSAN capable out of box.

40 hosts7,680 total vCPUs30.7 TB total RAM
Layer 2Network
100GbE leaf-spine fabric, non-blocking

4× spine switches (32-port 400GbE), 8× leaf switches (64-port 100GbE). All hosts connected at dual 100GbE. Separate out-of-band management fabric. ECMP routing, zero oversubscription at leaf-to-spine.

4 spine / 8 leaf100GbE per host (dual)Zero oversubscription
Layer 3Storage
All-flash Ceph-compatible storage pool

6× storage nodes, each with 12× 15.36 TB NVMe SSDs. 96 TB raw, ~55 TB usable (3× replication). 100GbE dedicated storage network (separate VLAN from compute traffic). S3-compatible object storage layer for backup and archive.

96 TB raw NVMe~55 TB usableS3-compatible object

Delivery in four phases

1Weeks 1–3Architecture design & BOM finalisation

Workload audit, capacity modelling, network topology design. BOM locked on week 3; PO issued same day.

2Weeks 4–8Hardware procurement & pre-staging

Compute nodes built and burn-tested. Switch firmware updated, port mappings pre-configured. Storage nodes assembled and tested in HK facility.

3Weeks 9–12Freight, customs & data-centre delivery

Air freight in bonded crates. Local customs agent coordinated by Haink. Hardware racked and cabled by Haink field team on arrival.

4Weeks 13–14Cluster commissioning & first VMs

Network fabric brought up, storage pool initialised, first 50 VMs migrated from foreign hyperscaler. Production sign-off on week 14.

Capacity at go-live

MetricDay-1 capacityPhase-2 expansion headroom
vCPU count7,680+3,840 (20 additional hosts)
RAM30.7 TB+15 TB
Usable storage55 TB+55 TB (6 additional storage nodes)
VM density (estimated)400 VMs800+ at full expansion
Network bandwidth (host)2× 100 GbESpine upgrade path to 400GbE

Related resources

Case Study AI-Ready Data Hall for a Telecom 400 kW, full BOM, 3-week delivery Case Study H100 Cluster for Gulf Sovereign AI 64 GPUs, InfiniBand, licensed export Hardware Server catalogue 2-socket EPYC, Xeon — all form factors Hardware Storage systems All-flash, hybrid and object storage

Planning sovereign or private cloud infrastructure?

Haink supplies full-stack compute, networking, and storage — designed together, delivered under one contract, no coordination gap between vendors.

sales@haink.org