InferaStack

The full stack from prompt to power

Sovereign AI infrastructure for Australia

InferaStack connects distributed compute, energy and network infrastructure into sovereign AI capacity. InferaGrid selects eligible resources across qualified sites; the Gateway provides the workload interface.

InferaGrid statusProof of concept · technical validation

Distributed infrastructureCompute · energy · network · sites
InferaGridQualify · admit · scheduleProof of concept · validation
Gateway + AI workloadsRequest · policy · record
Programs, relationships & credentials
NVIDIA Inception ProgramNEXTDC Partner Program ParticipantMSI Hardware Channel PartnerBytePlus Channel PartnerLumina Channel PartnerAWS Partner Network Member (Enrolled)ConnectVPP Energy Integration PartnerIncubated within the ACIA ecosystem

InferaStack Project Portfolio

CompanyInferaStack
Distributed AI InfrastructureInferaGrid
+
Sovereign ColocationData-Centre Deployments
+
OrchestrationInferaStack Gateway
+
AI ServicesModels · Workloads · APIs

InferaStack is the company and umbrella brand. InferaGrid is its distributed network; sovereign colocation, the Gateway and applied AI are complementary project lines.

PoC
Current InferaGrid stage
5
Validation gates before scale
0
Hardware-validated nodes
3 steps
Bench → node → first sites

What We Build

Distributed compute, dedicated GPU infrastructure, orchestration and applied AI.

InferaGrid · Proof of Concept

A sovereign orchestration network for distributed AI infrastructure

InferaGrid evaluates compute, power, network and location signals across qualified distributed sites. The reference architecture is at proof-of-concept stage.

  • Compute stays inside site power limits; safety and owner policy come first
  • Reserved capacity for always-on agents and scheduled AI work
  • Execution region and power-envelope state recorded per request
  • Validation gates: bench, one node, first site, repeatability, then decision
Built For
Site HostsEnergy & VPP OperatorsResearch OrganisationsSMEs & AI Product Teams
Sovereign AI Infrastructure & Colocation

AI deployment designed for NEXTDC's Tier IV network

Dedicated GPU environments in Australian data centres. InferaStack designs and commissions the deployment; you own the hardware. NEXTDC provides the facility, certifications and interconnect.

  • Site, power, capacity and rack design
  • Cooling, thermal engineering and acceptance tests
  • Private cloud on-ramps through NEXTDC AXON
  • Dedicated environments for regulated data
Built For
EnterprisesGovernment, Councils & Health ServicesResearch OrganisationsSMEs & AI Product Teams
Service Layer · In Development

The InferaStack Gateway

One OpenAI-compatible service layer for InferaGrid and data-centre deployments. It meters requests, enforces budgets and stores metadata-only audit records. The software is deployed in our AWS account for development but is not released.

Developers can preview the API surface at /docs and request early access.

Applied AI · Delivered for ACIA

InferaCard — the multilingual AI business card

InferaCard is ACIA's live multilingual AI business-card platform. ACIA owns and ships the product; InferaStack provides the AI for bilingual bios, professional headshots and intro reels.

AI services in practiceExplore InferaCard →

Why InferaStack

Australia needs more AI capacity without losing control of data or power. InferaStack combines distributed energy sites, certified data centres and one orchestration layer.

Energy-aware

Compute follows measured site load, solar, battery state and grid limits. Safety and owner policy come first.

Sovereign and auditable

Record the execution region per request. Store audit metadata, not prompts or completions.

Client-controlled

You own the hardware and choose the models. One OpenAI-compatible API reduces lock-in.

Gated, not hyped

Validate on the bench, then one node, then a few sites. Scale numbers remain scenarios until each gate passes.


Who It's For

For teams that need predictable GPU access and control over sensitive data. Site hosts and energy operators provide the supply side.

Research Organisations

Predictable GPU capacity under your governance

Reserve capacity for training, inference and data-heavy research without building a full on-campus cluster.

  • Reserved capacity at a fixed budget, through one OpenAI-compatible API
  • Private deployment when data must stay in your environment
Government, Councils & Health Services

Sovereign by design, recorded per request

Keep identified personal or clinical data in your own environment or a certified Australian facility. Each request records where it ran.

  • Identified personal or clinical data belongs in your own environment or a certified facility — by design, never on a residential node
  • Metadata-only audit records — prompts and completions are never stored
SMEs & AI Product Teams

AI on your own data, without handing it to a model provider

Run assistants and always-on agents over your own records, contracts and IP with predictable capacity and per-key budgets.

  • Dedicated or shared endpoints; GPU VMs; managed workers
  • Per-request metering, per-key budgets

How to Work With Us

Three ways in. No published price list yet — each deployment or reservation is scoped and quoted.

Colocation & Deployment
Project

A dedicated GPU environment in an Australian data centre, designed and commissioned by InferaStack. Quoted per deployment, from assessment to acceptance testing.

  • Assessment → design → acceptance
  • NEXTDC Partner Program participant
  • You own the hardware
Gateway Early Access
Preview

Preview the OpenAI-compatible API surface and join the early-access list for the hosted service layer.

  • API preview at /docs
  • Per-key budgets, metering
  • Not yet available to customers

Contact us to discuss the right starting point.


Our Roadmap

Design first. Then one node. Then sites. Progress by acceptance results and paid demand, not by calendar.

Phase 1 — Now (Stage 0a · Gate 0 pending)

Bench validation

The reference architecture is on paper. Next: test four GPUs and the admission, drain and recovery logic against a simulated power envelope.

Phase 2 — Next (Gates 1–2)

One node, then one site

Run one OEM-supported 8-GPU node for 72 hours, then qualify one approved field site after electrical, thermal, acoustic and energy-control acceptance.

Phase 3 — Gates 3–4

Repeat, then decide

Prove repeatability across a small set of qualified sites, then take the commercial decision on measured evidence and contracted demand. Larger figures remain scenarios.


Build or Reserve Sovereign AI Capacity

Tell us what you need. We will explain the current stage and the next practical step.

Plan an AI deployment

Talk to Us