CANUS
PortaDC vault and V-bank dry cooler on a substation pad at dusk

PORTADC™ · POWERED BY CANUS

PortaDCPOWERED BY CANUS

The AI factory that ships.450 kW to 1.2 MW per pod.

450 kW to 1.2 MW liquid-cooled AI pods, in 20 ft (PortaDC20 · 450 kW IT) and 40 ft (PortaDC40 · 900 kW to 1.2 MW IT) two-module stacks, owned by CANUS and leased to you by the IT-kW on sites that already have power and fibre. 12 to 21% under Sydney metro all-in. Zero evaporative water. 120-day target deployment from order.

SYS // 01 · Legacy data centre vs PortaDC

The concrete hall was built for a different decade.

  1. 1/6th

    Build cost per MW

    Pod and roof module against a metro concrete hall, before civil, crane and connection works. Factory-built to CANUS specifications and pre-certified to AS/NZS standards prior to site delivery.

  2. 120 days

    Target to powered racks

    120-day target deployment from order. Sites that already have power and fibre: solar-and-battery farms and industrial premises. No metro grid queue.

  3. 0 L

    Evaporative water after fill

    Closed-loop liquid cooling, heat rejected to air. Supply at or below 35 °C at a 45 °C design ambient, 50 °C survival margin.

  4. 12–21%

    Under Sydney metro, all-in

    Pulse rent plus power at cost against about US$270 in a Sydney hall: about US$238 at full 24/7 draw (12% under), about US$213 at a 65% average IT draw typical of inference fleets (21% under).

SYS // 02 · Two-module stacked architecture · Two hulls

450 kW to 1.2 MW per pod. Two purpose-built hulls.

Every PortaDC ships complete: compute below, heat rejection above, one two-module stack rated for 45 °C Australian summers with no evaporative water. One standard 450 kW building block: a single block in the compact 20 ft PortaDC20, two blocks in the 40 ft PortaDC40 (900 kW IT standard, up to 1.2 MW).

PortaDC20 two-module stack

20-foot high-cube ISO two-module stack

PortaDC20

450 kW IT · ~545 kW facility input · compact high-density footprint

Built for 3 to 5 high-density direct-to-chip racks (80 to 130+ kW/rack)

When your compute is liquid-cooled at 80 to 130+ kW per rack, 450 kW of AI capacity packs into 3 to 5 cabinets. PortaDC20 houses those racks with redundant liquid cooling, precision air trim and an Australian-certified switchboard inside a single 20 ft vault, with the heat-rejection module locked on top.

Short interconnect runs. Two lifts onto a ready pad.

Dimensions
6,058 × 2,438 × 2,896 mm lower vault + matching 6,058 mm rooftop heat-rejection module
Rack capacity
3 to 5 direct-to-chip liquid racks at 80–130+ kW/rack (+1 network/control bay)
Ideal hardware
NVIDIA GB200/GB300 NVL72, dense HGX B200/B300 & H200 liquid clusters
Cooling plant
Redundant N+1 liquid cooling · thermal buffer · precision air trim
Electrical
One 415 V AC bay · ties into existing solar, BESS or industrial switchboards
Logistics
Truck delivery · two-module lift and lock on a ready pad
PortaDC40 two-module stack

40-foot high-cube ISO two-module stack

PortaDC40

900 kW IT standard (up to 1.2 MW max) · ~1.09 MW facility input · dual-zone scale hull

Built for 8 to 12 liquid, hybrid or storage-rich racks (900 kW IT standard on dual 415 V bays, or 450 kW single-bay mode)

With a 12.2-metre lower vault and a full-length rooftop heat-rejection deck, PortaDC40 doubles the thermal and electrical capacity of the 20 ft pod. It houses two independent 450 kW power and liquid-cooling zones (900 kW IT standard, scalable to 1.2 MW IT) across 8 to 12 racks.

Tenants deploy full multi-rack GPU clusters alongside dedicated petabyte NVMe RAG and checkpoint storage racks inside a single weatherproof unit.

Dimensions
12,196 × 2,438 × 2,896 mm lower vault + matching 12,196 mm rooftop heat-rejection module
Rack capacity
8 to 12 racks at 65–120 kW/rack (900 kW–1.2 MW liquid mode) or 35–65 kW/rack (hybrid/storage mode)
Ideal hardware
Multi-rack NVIDIA GB200/GB300, HGX B200/H200 clusters, RTX PRO 6000 inference fleets + petabyte NVMe storage racks
Cooling plant
Dual redundant N+1 liquid cooling (1.0–1.2 MW thermal) · thermal buffer · hybrid air trim
Electrical
Dual 415 V AC bays (900 kW IT standard / 1.2 MW max) · configurable for single-bay 450 kW sites
Logistics
Semi-trailer or crane delivery · two-module vertical stack

SYS // 03 · Three purpose-plumbed pods

Pulse. Surge. Forge.

Two two-module hulls, PortaDC20 (450 kW IT) and PortaDC40 (900 kW to 1.2 MW IT), leased by the IT-kW across three operating envelopes. Each lease is defined by a firm floor, a cushion and a flexible solar band. Rents are indicative and confirmed in each lease.

PortaDC Pulse™ in PortaDC20

The 24/7 live inference & decode pod

Firm floor plus cushion. Flexible band: zero. Never trimmed.

Production inference needs a floor nobody can move. Pulse tenants run their whole allocation on the firm floor, with the cushion underneath it; the host's flexibility comes from the cushion, never from the tenant's racks. Air-cooled and hybrid racks slide in unmodified.

1.10 – 1.21 Design PUE

Two cooling configurations, one hull

  • 1AHybrid: unmodified air-cooled serversLiquid-cooled rear-door cabinets take the rack exhaust. In-row air covers the storage, optics and power supplies the doors do not reach.
  • 1BDirect-to-chip: liquid-ready racksCold plates on every GPU and CPU, rejected through the rooftop dry cooler. 1.10 – 1.21 design PUE in either configuration.

What it houses

Hull configuration

PortaDC20 · 450 kW IT (~545 kW facility input, 1 × 415 V bay, 3 to 5 racks)

Hardware & rack breakdown

4 × high-density liquid-cooled inference racks (HGX H200 / B200 or GB200) + 1 × low-latency optical gateway & hot-tier NVMe cache rack.

How it optimises this workload

Packs the entire 450 kW decode cluster into a 6-metre footprint, so every GPU sits within short copper interconnect reach, cutting inter-rack latency and time-to-first-token on 100% firm power.

24/7 firm production inference

US$165–172 (A$235–245) per IT-kW-month

Target rent, power passed through at cost. About US$213 to 238 all-in against Sydney metro at about US$270, 12 to 21% under. Flexible band: zero. Pulse tenants are never trimmed.

Workloads

  1. Real-time token streaming

    Low-latency serving around the clock on a firm floor the host cannot cut. Your models, your stack, your endpoints.

  2. Live voice, support and clinical agents

    24/7 low-latency endpoints for Australian banks, telehealth and enterprise software.

  3. On-demand image and video APIs

    Diffusion and video serving with weights kept warm in local storage.

The thesis

Move the data centre to the electrons.

Every sunny day across the NEM, utility-scale Solar + BESS farms spill energy the line cannot carry, or sell it at the solar capture price, while in Sydney and Melbourne AI companies wait for halls and pay about US$270 per IT-kW-month all-in. Instead of hauling expensive electricity down congested lines into a city building that costs about six times as much per megawatt, CANUS drops the pod straight onto the connected pad and sends the tokens back down fibre.

SYS // 04 · FrontierGate™ + LARUS AI

The brain inside every pod.

A brushed-gunmetal unit, patent pending, in every pod. FrontierGate™ enforces strict electrical isolation between the host asset and the compute vault in sub-millisecond windows, so the tenant's firm floor never moves. LARUS AI™ signs boundary power and thermal delivery into the operating record a lender, a host and a tenant read.

FrontierGate boundary governor with site load waveforms

  1. 01 // The cushion, executed at the boundary

    5:00 PM host load spike

    The host's own load ramps toward the approved boundary. FrontierGate™ holds the host's power boundary while the tenant's firm floor never moves. A Surge tenant receives the flexible-band signal and answers with its own orchestration; a Pulse tenant feels nothing.

    • Host boundary: held
    • Tenant firm floor: unchanged
    • Pulse racks trimmed: never

  2. 02 // The buffer carries the cushion

    45 °C summer afternoon

    At the 45 °C design ambient the pod holds supply coolant at or below 35 °C through a host call, with a 50 °C survival margin above that. The host gets its flexibility; the tenant never feels it.

    • Supply coolant: at or below 35 °C
    • Design ambient: 45 °C · survival 50 °C
    • Evaporative water: none after fill

  3. 03 // Two channels, hardware-signed

    The LARUS AI operating record

    LARUS AI™ cryptographically signs boundary power and thermal delivery into a single auditable ledger, alongside independent hardware fault isolation. Two parties read the one record: the tenant for its invoice and SLA, the host for its boundary and settlement.

    • Boundary power // Hardware-signed
    • Thermal delivery // Hardware-signed
    • Fault isolation // Independent, hardware

SYS // 05 · Four layers

Designed for 45 °C. Survives 50 °C.

  1. 01

    Fenced, monitored switchyard perimeter

    Sited inside fenced, 24/7 monitored solar-and-battery switchyards or secured industrial premises. No public foot traffic.

  2. 02

    Windowless weathering-steel armour (IP55)

    Solid corrugated weathering-steel walls with no air-intake louvres, internal deadbolts, vibration sensing and clean-agent fire detection and suppression to Australian standards.

  3. 03

    One signed record, one fault trip

    LARUS AI™ hardware-signs boundary power and thermal delivery into one auditable ledger. Tamper-linked telemetry and independent hardware fault isolation sit alongside.

  4. 04

    Factory-tested. Certified in Australia.

    Every pod passes witnessed factory acceptance testing before it ships, then Australian integration and certification: AS/NZS 61439 switchboard, revenue-grade boundary metering, fire detection and suppression to Australian standards, statutory AS/NZS 3000 certification and site acceptance sign-off.

IR-Stealth™ · Thermal armor

Looks like tactical gunmetal to the eye.Behaves like reflective white to the Outback sun.

Conventional dark enclosures absorb most of the sun's radiation. Every PortaDC vault is finished in a marine-grade, low-solar-absorbing thermal armor that reflects the near-infrared the eye cannot see, so the skin stays cool in 45 °C desert sun and the cooling plant spends its capacity on GPU heat, not sun load. Marine-grade corrosion protection and no glare on BESS and solar sites.

Near-infrared reflected
700–2,500 nm
Desert sun, cool skin
45 °C
Marine corrosion class
C5-M
Glare on site
Zero

Near-infrared, 700 to 2,500 nm

  1. 01Low-solar-absorbing reflective topcoatNaval grade · gunmetal to the eye
  2. 02Ceramic thermal barrierThe mirror under the colour
  3. 03Fire-rated insulated coreClass A

Legacy metro data centre vs PortaDC

Six numbers that decide the build.

MetricLegacy metro data centre (Sydney / Melbourne)PortaDC by CANUS (Solar + BESS substation or on-site)
Build cost per MW (excl. GPUs)

About 6× the pod

Concrete halls, chillers, land

About 1/6th of a metro hall

Factory-built pod on a ready pad

Time to powered racks

Years

Metro grid queues

120 days

Target deployment from order, on existing 415 V boards

Power

Metro tariff

Retail grid mix plus network charges

Behind-the-meter renewable and hybrid tariffs

Passed through at cost under transparent site schedules (shared under NDA)

Rack density supported

10 – 25 kW / rack

Legacy air cooling

35 – 130+ kW / rack

Forge 80 – 130 kW, direct-to-chip liquid

Water consumption

Millions of litres / year

Evaporative cooling towers

No evaporative water after fill

Closed-loop liquid, rooftop dry cooler

All-in monthly cost

~US$270 / IT-kW / month

Sydney metro, all-in

~US$213 / IT-kW / month

Pulse target rent + power at cost · 12 to 21% under

SYS // 06 · Three leases, one pod

Bring your GPUs. Lease the capacity.

Target rent per IT-kW-month, power passed through at cost · US$ first, A$ beside

PortaDC Surge™

Elastic batch and prefill. 25 to 40% contracted flex, tenant-run.

Rent, US$ (A$)
140 – 155 (A$200 – 220)
All-in with power, US$
~182
vs Sydney metro US$270
33% under

PortaDC Pulse™

24/7 firm production inference. Flexible band: zero.

Rent, US$ (A$)
165 – 172 (A$235 – 245)
All-in with power, US$
~213
vs Sydney metro US$270
12 to 21% under

PortaDC Forge™

Dense training. Firm liquid racks at 80 to 130 kW.

Rent, US$ (A$)
180 – 187 (A$255 – 265)
All-in with power, US$
~236
vs dense hall US$311
24% under

Indicative figures, confirmed in each lease. Power figures are planning blends, not a host tariff. Enquiries do not quote prices or reserve inventory.

Engineered for & compatible with

NVIDIA GB200 / H200 / RTX PRO 6000AMD InstinctDell PowerEdgeHPE CraySupermicroLenovo NeptuneGigabyteAristaBroadcomAmphenolTE ConnectivitySolidigm NVMeNVIDIA DynamovLLM

THIRD-PARTY BRANDS AND MODELS IDENTIFY COMPATIBLE HARDWARE, INTERFACES, AND SPECIFICATION STANDARDS ONLY. NO AFFILIATION, ENDORSEMENT, OR PARTNERSHIP IS IMPLIED.

Reserve capacity or host a pod

Deploy a pod.

Three ways in. Pick your track; a named person replies within one business day. Enquiries do not quote prices or reserve inventory.