Platform · Coming soon
Lumus platform

Dedicated compute. A path to managed inference.

Australian AI infrastructure starting with a 144-node NVIDIA B300 cluster in Sydney, with a path towards GPU, token and agent services.

First cluster · Sydney
144 × NVIDIA B300 nodes
Coming soon
Managed inference
API request
Coming soon
POST /inference
{
  "model": "<selected-model>",
  "input": "<your-prompt>"
}

// Interface and endpoint TBC
First cluster
144 nodes · 1,152 GPUs
Platform
NVIDIA B300
Initial consumption
Reserved capacity, bare metal
Future service layer
Managed inference remains TBC
Infrastructure

Our first cluster. Built in Sydney.

The first Lumus cluster brings 144 NVIDIA B300 nodes to Sydney, built for reserved capacity and inference-led workloads. Further sites and configurations will be announced as they are confirmed.

01 / Sydney · B300 infrastructure Coming soon

144-node NVIDIA B300 cluster

A first Australian cluster shaped around reserved capacity and predominantly inference-led demand.

  • Cluster size144 nodes
  • PlatformNVIDIA B300
  • Commercial starting pointReserved capacity
  • Operating pathManaged mobilisation
Cluster map

Visualising 144 nodes.

Each tile is one NVIDIA B300 node in the first Sydney cluster, giving the scale of the platform a clear, legible form.

Simplified view. Networking, storage and facility details are not shown.
Inference API · Coming soon

A future path from GPU capacity to managed model access.

The Lumus product ladder can move from reserved bare metal to GPU as a service, token as a service and eventually agent systems. This interface explores how that future offer could be presented.

Lumus managed inference
Coming soon
Language

General language model

Model catalogue and licensing to be confirmed.

Per-token modelTBC
Coming soon
Reasoning

Reasoning model

Model catalogue and licensing to be confirmed.

Per-token modelTBC
Coming soon
Multimodal

Vision-language model

Model catalogue and licensing to be confirmed.

Managed accessTBC
Coming soon
Language

Compact language model

Model catalogue and licensing to be confirmed.

Per-token modelTBC
Coming soon
Embedding

Embedding model

Model catalogue and licensing to be confirmed.

Managed accessTBC
Coming soon
Agent systems

Agent service layer

Portal, workflow and service design remain future scope.

Portal modelFuture
Future service
01

Control plane

A managed or shared service needs a stronger platform layer than reserved bare metal.

02

Isolation

Customer and workload boundaries must be designed into the shared service model.

03

Metering

Token-based consumption requires measurable usage and a defined commercial model.

04

Lifecycle

Models and the serving layer need an ongoing operational and support boundary.

Platform evolution

Start with control. Add services with capability.

Lumus is building from dedicated infrastructure towards increasingly managed AI services, giving customers more ways to access compute as the platform grows.

01
Initial focus

Bare metal and reserved capacity

Customers bring their own software environment and consume dedicated capacity.

02
Next service layer

GPU as a service

GPU capacity is combined with a platform layer and a clearer operational boundary.

03
Future · TBC

Token as a service

Lumus runs selected models and meters customer consumption by token.

04
Future service

Agent as a service

Complete agent systems are hosted within the Lumus platform and accessed through a portal.

Shape the platform

Planning an inference-led workload?

Talk with Lumus about reserved capacity requirements or register interest in the upcoming managed inference service.

Start a conversation