Infrastructure

One reference design. Three places to run it.

Every IntelligentX deployment uses the same validated GPU and CPU design and the same software stack. What changes is the building: ours, yours, or your cloud provider's.

NVIDIAAMDDellSupermicro
IntelligentX CloudYour data centerYour cloud
Who owns the hardwareIntelligentXYou (buy or finance)Your cloud provider
Who runs itIntelligentX, in-countryIntelligentX remotely, or you, air-gappedIntelligentX, in your tenancy
Capital outlayNoneHardwareNone; uses your committed spend
Air-gappedNoYesNo
Data leaves your controlNever leaves the countryNever leaves your siteNever leaves your account
PIES, Zetaris, private models, governanceIncludedIncludedIncluded

Reference design

Sized for inference, apps and data

Enterprise AI is mostly inference, applications and databases, with fine-tuning in bursts. The same design goes into our regions and onto your floor; in your cloud it maps to equivalent GPU instances.

REFERENCE RACK · OUR REGION OR YOUR DATA CENTER · 42U
GPU · large modelsNVIDIA HGX B200

8-GPU nodes on Dell PowerEdge or Supermicro for LoRA fine-tuning and the Pro, Expert and Ultra private models.

GPU · dense inferenceNVIDIA RTX PRO 6000 Blackwell

Air-cooled servers serving the Lite and Dev models at high density. Fits standard data-center power and cooling.

CPU · apps and dataAMD EPYC

Kubernetes on high-core-count EPYC servers runs PIES Studio, Zetaris, every generated application and its database.

StorageAll-flash NVMe

Model weights, datasets and application data, with encrypted snapshots kept in the same jurisdiction.

NetworkIsolated tenant fabric

Per-tenant isolation and no public egress by default. Private interconnect to your sites.

Cloud mappingAWS and Azure GPU instances

In your cloud, the design maps to equivalent NVIDIA GPU instances in your chosen region.

AUSTRALIA · FIRSTSYD1 · Sydneyfirst region · enterpriseCBR1 · Canberragovernment · protectedopening first UNITED STATES · NEXTDFW1 · Dallasregulated enterpriseIAD1 · N. Virginiafederal · financenext EUROPE · THENFRA1 · FrankfurtEU data residency · AI Actthen Sprints start immediately on in-country sovereign capacityand move onto IntelligentX regions as they open. Each tenant is pinned to one jurisdiction.

IntelligentX regions

Region by region, each tenant pinned to one jurisdiction

Australia. Sydney opens first, close to existing PIES customers, followed by Canberra for government. United States. Dallas–Fort Worth, then Northern Virginia for federal and financial services. Europe. Frankfurt, for EU data residency and the AI Act.

Each region is colocated in an existing Tier III or IV facility. Customers who need to start sooner choose their own data center or cloud and move later, or stay.

Operations

What a customer gets, wherever it runs

  • 01
    Dedicated GPUNever shared with another tenant. Unlimited private inference with no per-token fee.
  • 02
    CPU and database capacityFor the applications PIES generates and the Zetaris data layer behind them. Sized in the sprint, grown with use.
  • 03
    One operatorIntelligentX monitors, patches and upgrades the whole stack, from firmware to applications, in every option except fully air-gapped.
  • 04
    Compliance built for the buyerEach region is designed to the buyer's framework, IRAP in Australia, SOC 2 and HIPAA in the United States, GDPR and the EU AI Act in Europe, with independent assessment on our roadmap. Air-gapped where required.

Start

Not sure which option fits?

The first week of every sprint answers it, from your data, your regulator and your existing commitments.

Talk to us →