Skip to main content

Infrastructure that stays under control.

Servers, networks, GPU compute and security designed, deployed and documented for on-premise, cloud, hybrid and air-gapped environments.

01 Deployment models

On premises, in the cloud or disconnected.

The same engineering standards apply whichever model fits the security requirements.

  • On premises

    Where it runs
    Client data centre or server room
    Internet access
    Optional
    Updates
    Online or scheduled
    Suited to
    Full physical control

  • Cloud

    Where it runs
    Public or private cloud, or hybrid
    Internet access
    Controlled
    Updates
    Managed pipelines
    Suited to
    Elastic and distributed workloads

  • Fully disconnected

    Air-gapped

    Where it runs
    Isolated network, no external route
    Internet access
    None
    Updates
    Controlled transfer
    Suited to
    Classified and critical systems

02 Reference designs

Two reference designs, one standard.

A proven segmented network for smaller projects, and a cloud-native AI platform prepared for bigger ones. Both encrypted end to end and delivered as code.

Design A · Proven foundation

Segmented by purpose, verified by default.

Networks are segmented by purpose, every request is verified, every link is encrypted and the external link is optional. The same design runs connected or fully air-gapped.

Best for Smaller projects: a single site, a focused team, a stable set of systems.

Network topology: an optional external link, a perimeter firewall, core switching and isolated segments for users, compute, storage and management, with identity verified on every request. The design also runs fully air-gapped.

External networks

Encrypted

Air gap Updates by controlled transfer

Perimeter firewall

Encrypted

Core switching

  • VLAN 10

    Users and endpoints

  • VLAN 20

    Compute

  • VLAN 30

    Storage

  • VLAN 99

    Management

Zero trust across every segment.

Design B · Cloud-native AI platform

Private AI, built cloud-native.

Cloud-native means containers on Kubernetes clusters, abstracted from the hardware underneath and spread across sites and edge locations as one distributed platform. It still runs on owned or controlled compute, on premises or air-gapped.

Best for Bigger projects: distributed networks across sites and edge, many teams and models, workloads that grow.

Cloud-native AI platform in four layers: an AI gateway behind zero trust access; model serving from core to edge; Kubernetes clusters on two sites joined by a service mesh; and GPU compute, data and observability. Encrypted traffic passes through every layer in both directions.
  1. 01

    AI gateway

  2. 02

    Model serving

  3. 03

    Multi-cluster platform

  4. 04

    GPU, data and observability

Risks designed out

  • Risk

    AI needs compute that is not there yet

    GPU capacity, storage and networking sized for inference and 3D processing.

  • Risk

    Security bolted on after go-live

    Zero trust and segmentation designed in from the first drawing.

  • Risk

    Knowledge held by a few people

    Infrastructure as code, documentation and operating procedures handed over.

03 Build log

From survey to sign-off.

Each stage closes with a tangible output before the next opens, so the environment is never half built and undocumented.

dc.live
An aisle of server racks in a data centre

  1. 01 Assess

    Output · Site survey

  2. 02 Design

    Output · Reference design

  3. 03 Procure

    Output · Bill of materials

  4. 04 Install

    Output · Commissioned systems

  5. 05 Harden

    Output · Test records

  6. 06 Hand over

    Output · Operations manual

Planning new infrastructure?

Describe the environment, the workloads and the constraints. The answer starts with an architecture, not a product list.

Discuss the infrastructure